Jurnal Teknik Informatika (Jutif)

ANALYSIS OF THE MOVIE DATABASE FILM RATING PREDICTION WITH ENSEMBLE LEARNING USING RANDOM FOREST REGRESSION METHOD

2025-02-11T00:25:37+00:00

The film industry has become a very profitable industry. However, during COVID-19 the film industry experienced an unfavorable impact with the delay in the screening schedule of new films, many cinemas were prohibited from operating so they were completely closed, and it wasn’t easy to obtain permits to carry out the filmmaking process. To survive in this industry from the impact of the pandemic, it is necessary to consider several factors such as targeted promotion methods by using the right selection of predictive decisions with market and trends. Predicting the success of a film is very helpful in determining the success rating and quality of the film to be released. The Random Forest Regression method is used to conduct predictive analysis on films. This study uses the M-estimate encoding technique to handle categorical data into numerical data, and the result shows that the application of M-estimate encoding increases the correlation value between features. In the Random Forest Regression method with 1000 trees, dividing 80% training data and 20% testing data, the R2 performance score was 86%, the MSE score was 12%, the RMSE score was 35% and the MAE score was 22%. The 10-fold cross-validation score in this study was 85%. This shows that the Random Forest Regression method using 80% training data produces the best performance score.

AN ENHANCED MULTI-LAYERED IMAGE ENCRYPTION SCHEME USING 2D HYPERCHAOTIC CROSS-SYSTEM AND LOGISTIC MAP WITH ROUTE TRANSPOSITION

2025-02-11T00:25:42+00:00

In the rapidly evolving digital era, image encryption has become a crucial technique to protect visual data from the threat of information leakage. However, the main challenge in image encryption is improving security against cryptanalysis attacks, such as brute-force and differential attacks, which can compromise the integrity of the encrypted image. Additionally, the creation of efficient and fast encryption schemes that do not degrade image quality remains a significant challenge. This research proposes a multi-layer image encryption scheme that integrates the Logistic Map algorithm, Cross 2D Hyperchaotic (C2HM) system, and Route Transposition techniques. The method aims to enhance the security of digital image encryption by combining chaotic and hyperchaotic systems. The Logistic Map is used to generate a sequence of random values with high chaotic properties, while C2HM contributes to increasing complexity and variability. The Route Transposition technique is applied to scramble pixel positions, further strengthening the encryption’s randomness. The encryption key is derived from a combination of the image hash and user key, which are then used to calculate the initial seed in the chaotic algorithm. Experiments were conducted using standard images with a resolution of 512×512 pixels. The security analysis includes evaluations of NPCR, UACI, histogram analysis, and information entropy. The experimental results show that NPCR consistently exceeds 99.5%, while UACI ranges between 33.23% and 33.56%, indicating high sensitivity to minor changes. Histogram analysis demonstrates an even intensity distribution, and the information entropy value of 7.999 reflects an exceptionally high level of randomness. Robustness tests also indicate that this method can maintain image integrity even when subjected to damage or data loss.

TEXT CLASSIFICATION OF BULLYING REPORTS USING NLP AND RANDOM FOREST.

2025-02-11T00:25:48+00:00

Bullying is a great concern that needs to be dealt with as early as possible, be it in the form of physical, verbal, social or cyber bullying. Using NLP algorithms, this paper intends to classify bullying report using Natural Language Processing in conjunction with Bag of Words. The study employs quantitative methodology. A total of 4671 reports of bullying are in essence categorized into physical, verbal, social, cyber and non-cyber bullying. We split the dataset into 80% training set (3737 reports) and 20% testing set (934 reports). The above model has achieved an accuracy of 94,76%, with good values of recall, precision and F1-score: 94,64%, 95,02% and 94,97% respectively. The dataset is then analyzed using Random Forest algorithm and Report of the Bullying Survey The model is to be effective in automatic Detection of Textual Bullying Reports Automated. While there has been no such effort in our institutions so far, automatic reporting of bullying will prove to be effective. This is because the system will allow a school or institution to have a precise constant monitoring of bullying reports. It will also allow an instantaneous action to be taken to protect the victim without letting the situation escalate.

DESIGN AND IMPLEMENTATION OF A METAVERSE-BASED PLATFORM FOR THEMATIC TOURISM VILLAGES

2025-02-12T11:32:11+00:00

This research investigates the integration of metaverse technology into the design of thematic tourism villages to address the declining appeal of such destinations and the limited adoption of digital tools for promotion and education. Using the Kayutangan Heritage Village in Malang as a case study, the study employs a Design Thinking approach to identify challenges, ideate solutions, and develop a metaverse-based platform prototype. This platform leverages immersive 3D environments to enable users to explore cultural heritage and historical narratives in a virtual context. The findings highlight the platform's ability to create resilient, engaging, and educational tourism experiences that attract diverse audiences while supporting local economies. By offering innovative, interactive ways to connect with cultural heritage, the metaverse platform demonstrates its potential as a transformative tool for digital tourism. The research contributes to the broader discourse on sustainable tourism by showcasing the metaverse’s role in preserving cultural heritage and enhancing digital connectivity. Its impact is evident in fostering a deeper understanding of cultural assets and promoting digital tourism innovations that align with contemporary market demands, thus paving the way for the widespread adoption of metaverse technology in similar contexts.

SENTIMENT ANALYSIS OF INDONESIA'S CAPITAL RELOCATION USING WORD2VEC AND LONG SHORT-TERM MEMORY METHOD

2025-02-12T11:47:25+00:00

The relocation of the national capital (IKN) has garnered public attention, triggering various reactions and sentiments among the community. Sentiment analysis is crucial for understanding public perceptions of an issue, particularly on social media platforms like Twitter and YouTube. This study's sentiment analysis employs Word2Vec parameters, including architecture and dimensions. Additionally, hyperparameters such as the Optimizer and activation functions are applied to the Long Short-Term Memory (LSTM) model to analyze their effect on sentiment classification performance related to the IKN relocation. The study aims to compare the influence of Word2Vec parameters on LSTM model hyperparameter performance in sentiment classification. Data on the IKN relocation were gathered from tweets and YouTube video comments, then processed to form a text corpus used to train the Word2Vec model with Skip-gram and Continuous Bag-of-Words (CBOW) architectures, utilizing different dimension sizes (100 and 300) to enhance word representation in vectors. After obtaining word representations, the LSTM model was applied to classify sentiments using hyperparameters such as activation functions (ReLU, Sigmoid, and Tanh) and two Optimizers (Adam and RMSProp). The results indicate that the Skip-gram architecture tends to yield higher accuracy compared to CBOW, particularly with larger vector dimensions (300), which generally improved model accuracy, especially when using the RMSProp Optimizer and ReLU activation function, achieving an accuracy of 91%. It can be concluded that dimension values and architecture in Word2Vec, as well as the use of Optimizer and activation functions in LSTM, significantly impact model performance.

OPTIMIZATION OF STOCK PRICE PREDICTION WITH RIDGE REGRESSION AND HYPERPARAMETER SELECTIONS

2025-02-12T11:43:00+00:00

Stock price prediction is a topic that has garnered significant attention in the investment world and has been the subject of various studies. Despite the massive attention, predicting stock price movements using algorithms remains challenging as the algorithms must be agile and highly adaptive to movement trends. Recent studies using deep learning methods for stock price prediction show that deep learning methods have high reliability. However, their computational complexity limits widespread implementation. This study aims to predict Netflix stock prices using a linear regression model with ridge and hyperparameter optimisation. The research consists of three stages: data preprocessing, building a linear regression model with ridge, and predicting and visualizing results. The dataset used is historical Netflix stock price data from 2017 to 2022. In the preprocessing stage, the data was normalized using MinMaxScaler and split into training and test sets. A ridge regression model was built with hyperparameter alpha optimization using GridSearch. Predictions were compared to stock prices and evaluated using Root Mean Squared Error (RMSE). The ridge regression model with hyperparameter optimization performed best with an RMSE of 13.8082. Although the linear regression model demonstrated the fastest execution time of 0.7717 seconds, the ridge regression model with hyperparameter optimization provided an optimal balance between prediction accuracy and time efficiency.

PERFORMANCE COMPARISON OF NAIVE BAYES AND BIDIRECTIONAL LSTM ALGORITHMS IN BSI MOBILE REVIEW SENTIMENT ANALYSIS

2025-02-12T11:50:34+00:00

Currently, almost all banks have used mobile banking in conducting banking transactions, one of which is Bank Syariah Indonesia (BSI). BSI mobile is still classified as a new mobile banking application compared to other mobile banking, this certainly still has a low rating and really needs feedback from users which can be seen through reviews on the Google Play Store application. Input in the form of criticism and suggestions from BSI mobile users can be used by BSI mobile as a suggestion for careful supervision and evaluation material in improving its services. This study aims to find the best algorithm to analyze review sentiment on the Google Play Store for the BSI mobile application and provide an overview of the response of application users to application developers based on the results of review data processing. The data mining methodology used in this study is CRISP-DM, using a dataset collected for 6 years (2018-2023) which is annotated into positive and negative labels manually, then modeled using 2 algorithms, namely Naïve Bayes (NB) and Bidirectional LSTM (BiLSTM). The contribution of this study is to test, evaluate and compare the two algorithms (NB and BiLSTM) using the K-Fold Cross Validation (NB) testing model and over-sampling techniques to the minority class (negative) then provide recommendations for the best algorithm. The conclusion of the study is that the BiLSTM algorithm is superior to NB with an accuracy of 94.90 % while the NB algorithm is 94%. In addition, the over-sampling technique is more optimal in increasing the accuracy of the algorithm's performance compared to without over-sampling.

PERFORMANCE COMPARISON OF NAIVE BAYES, SUPPORT VECTOR MACHINE AND RANDOM FOREST ALGORITHMS FOR APPLE VISION PRO SENTIMENT ANALYSIS

2025-02-12T10:55:32+00:00

With the development of spatial computing devices, there arises a need to analyze consumer opinions about products such as the Apple Vision Pro (AVP), a technology that combines augmented reality (AR) and virtual reality (VR). This study aims to analyze consumer opinions on the Apple Vision Pro by utilizing data from the social media platform X. Three algorithms—Random Forest, Support Vector Machine (SVM), and Naïve Bayes—are used in text categorization to identify sentiment trends. Data was collected through a crawling process, resulting in 3,753 entries. After preprocessing and labeling, 2,609 clean data points were obtained, with 1,618 classified as negative and 991 as positive. In sentiment analysis, Random Forest delivered the best performance with an accuracy of 83%, followed by SVM at 80%, and Naïve Bayes at 75%. These results indicate that the Random Forest algorithm is more effective in sentiment categorization related to Apple Vision Pro. This study provides significant contributions to companies in understanding public perceptions and crafting more precise data-driven marketing strategies.

DETECTION OF BULLYING CONTENT IN ONLINE NEWS USING A COMBINATION OF RoBERTa-BiLSTM

2025-02-12T11:06:28+00:00

This research aims to build a bullying-themed online news classification system with a combined approach of RoBERTa embedding and BiLSTM. RoBERTa is used to generate context-rich text representations, while BiLSTM captures temporal relationships between words, thereby improving classification performance. The research dataset consisted of news from reputable portals such as Kompas.com, Detik.com, and iNews.com, labeled according to keywords relevant to the theme of bullying. The results of the experiment showed that the model achieved 95.2% accuracy, 98.2% precision, 93.6% recall, and 95.8% F1-score. Although there are few prediction errors (false positives and false negatives), this model shows excellent performance in detecting and classifying bullying-themed news. The main contribution of this research is the development of a new approach that combines RoBERTa and BiLSTM for the classification of complex bullying-themed news. This approach not only improves the accuracy of classification but can also be implemented in automated systems to detect negative content. Thus, this research has the potential to support the creation of a healthier digital space and encourage more responsible media practices.

FOOTBALL PLAYER TRACKING, TEAM ASSIGNMENT, AND SPEED ESTIMATION USING YOLOV5 AND OPTICAL FLOW

2025-02-12T11:09:39+00:00

Football analysis is indispensable in improving team performance, developing strategy, and assessing the capabilities of players. A powerful system that combines YOLOv5 for object detection with optical flow tracks football players, assigns them to their respective teams, and estimates their speeds accurately. In the most crowded scenarios, the players and the ball are detected by YOLOv5 at 94.8% and 93.7% mAP, respectively. KMeans clustering based on jersey color assigns teams with 92.5% accuracy. Optical flow is estimating the speed with less than 2.3%. The perspective transformation using OpenCV improves trajectory and distance measurement, overcoming the challenges in overlapping players and changing camera angles. Experimental results underlined the system's reliability for capturing player speeds from 3 to 25 km/h and gave insight into the dynamic nature of team possession. However, there is still some challenge: 6% accuracy degradation in high overlap and illuminative changes. The future work involves expanding the dataset for higher robustness and ball tracking, which will comprehensively explain the dynamics of a match. The paper presents a flexible framework for automated football video analysis that paves the way for advanced sports analytics. This would also, in turn, enhance informed decision-making by coaches, analysts, and broadcasters by providing them with actionable metrics during training and competition. The proposed system joins the state-of-the-art YOLOv5 with optical flow and thereby forms the backbone of near-future football analysis.

DYNAMIC WEIGHT ALLOCATION IN MODIFIED MULTI-ATRIBUTIVE IDEAL-REAL COMPARATIVE ANALYSIS WITH SYMMETRY POINT FOR REAL-TIME DECISION SUPPORT

2025-02-12T11:19:56+00:00

Decision Support Systems (DSS) have a crucial role in real-time decision-making, especially in the digital era that demands high speed and accuracy. Managing criterion weights in a dynamic environment presents significant challenges due to rapid and unpredictable changes in conditions. However, determining an accurate weight becomes difficult due to uncertainty, incomplete data, and subjective factors from decision-makers. In addition, changes in the external environment, such as market trends, regulations, or customer preferences, can affect the relevance of each criterion, thus requiring a real-time weight adjustment mechanism. The purpose of this study is to develop and explore the dynamic weight allocation method in symmetry point- multi-attributive ideal-real comparative analysis (S-MAIRCA) to support more accurate and responsive real-time decision-making in a dynamic environment. This research contributes to the understanding of how the weights of criteria can be adjusted automatically and responsively to changing conditions or new data, which increases the relevance and accuracy of decisions in a dynamic environment. The urgency of S-MAIRCA research is important because it often involves real-time, dynamic, and complex data. This development not only improves the adaptability of the S-MAIRCA method, but also contributes significantly to creating computer science-based applications that are more intelligent, flexible, and relevant to the evolving needs of the system. The results of the alternative ranking comparison using the CRITIC-MAIRCA, LOPCOW-MAIRCA, ROC-MAIRCA, and S-MAIRCA methods showed variations in the ranking order generated for each alternative using spearman correlation. The results of the correlation value of CRITIC-MAIRCA and LOPCOW-MAIRCA have a very high correlation of 0.993, which shows that these two methods provide almost identical rankings in alternative evaluation. Likewise, CRITIC-MAIRCA and S-MAIRCA had a high correlation of 0.979, signaling a strong similarity in ranking results despite slight differences in the approaches used by the two methods. The results of the application of the MAIRCA-S method in the development of DSS based on real-time data have a significant impact on improving the speed, accuracy, and adaptability of decisions. MAIRCA-S strengthens the validity of decision results by considering a variety of attributes on a more comprehensive scale, providing added value in the development of DSS for various industrial sectors.

PERFORMANCE ANALYSIS OF EXTRACT, TRANSFORM, AND LOAD METHODS FOR BUSINESS INTELLIGENCE IN E-LEARNING SYSTEMS USING PENTAHO DATA INTEGRATION

2025-02-12T11:22:24+00:00

The rapid adoption of Learning Management Systems (LMS) in higher education has resulted in the generation of large and complex datasets, posing significant challenges for efficient data integration and analysis. The urgency to address these challenges is driven by the growing demand for real-time analytics and data-driven decision-making in educational institutions. This study advances the field of computer science by evaluating and comparing the performance of three Extract, Transform, and Load (ETL) methods—Table Output, Sync After Merge, and Switch Case—using Pentaho Data Integration (PDI). The study introduces novel insights into ETL optimization techniques, focusing on execution time as the primary metric, critical for ensuring timely and reliable insights in Business Intelligence (BI) systems. Performance testing was conducted with synthetic datasets ranging from 150 to 1,000,000 records across five scenarios: data addition, synchronization, insertion, deletion, and combined operations. Results reveal that Sync After Merge outperformed other methods, achieving up to 35% faster execution times, particularly with large datasets. These findings contribute significantly to the advancement of data integration techniques in computer science, enabling institutions to optimize their BI systems, enhance system responsiveness, and support data-driven decision-making processes effectively. The research provides valuable insights for developing scalable ETL solutions in educational technology systems.

EFFECTIVENESS HYPERPARAMETER TUNING ON RANDOM FOREST, LINEAR DISCRIMINANT ANALYSIS, LOGISTIC REGRESSION AND NAÏVE BAYES ALGORITHMS FOR DETECTING DOS NETWORK ATTACKS

2025-02-12T11:24:57+00:00

Denial of Service (DoS) attacks are a major threat to network security, characterized by overwhelming system resources with illegitimate requests. Such attacks can disrupt critical services and cause substantial financial losses. However, there is still a need for a more efficient model to detect DoS attack with high accuracy. The aim of this research is to determine the impact of hyperparameter tuning on the four algorithms to identify the best algorithm for detecting DoS network attacks. The research method involves data preprocessing, feature selection, encoding, balancing using SMOTE (Synthetic Minority Over-Sampling Techinuque) and evaluation using confusion matrix. This research use the NSL-KDD dataset because it is relevant for DoS attack detection and flexible for testing various classification algorithms and utilizing hyperparameter tuning. This study evaluates the effectiveness hyperparameter tuning on several machine learning alghorithms namely Random Forest, Linear Discriminant Analysis (LDA), Logistic Regression and Naïve Bayes in detecting DoS attacks. Results indicate that Random Forest achieves highest accuracy (99,97%) and robust performance across all metrics, demonstrating superior generalization and precision. LDA, Logistic Regression and Naïve Bayes also performed well but fell short of Random Forest in handling complex patterns in the dataset. The utilization of hyperparameter tuning can improve the accuracy, consistency and efficiency of the algorithm so as to optimize the combination of various parameters in detecting DoS attacks. The findings provide valuable insights into selecting suitable algorithms for future implementations in cybersecurity systems.

COMPARISON OF DBSCAN AND K-MEANS CLUSTER ANALYSIS WITH PATH-ANOVA IN CLUSTERING WASTE MANAGEMENT BEHAVIOUR PATTERNS

2025-02-12T11:31:05+00:00

This study aims to compare the effectiveness of DBSCAN and K-Means cluster analysis methods in clustering waste management behaviour patterns in Batu City. The data used is secondary data from previous research with a total of 395 respondents taken using the quota sampling method. DBSCAN classifies data based on density with the main parameters epsilon and MinPts, while K-Means uses the average centroid to determine the cluster. The analysis results show that DBSCAN produces a silhouette index of 0.664, which is higher than K-Means with a value of 0.574. DBSCAN also successfully identified noise as much as 10 data that did not belong to any cluster, while K-Means did not have a similar mechanism. The results of Path-ANOVA show that DBSCAN is the most optimal clustering with a more significant partition difference value. Further tests were conducted to strengthen the validation of clustering results using Path-ANOVA. Both methods produced two main clusters, with the second cluster showing higher quality in terms of maintenance, quality, and ease of use of environmental hygiene facilities. This research emphasises the importance of choosing an appropriate clustering method to ensure optimal clustering results, especially in data with complex characteristics.

MODELING INTRUSION DETECTION AND PREVENTION SYSTEM TO DETECT AND PREVENT NETWORK ATTACKS USING WAZUH

2025-02-12T11:53:47+00:00

The rapid development of technology has a positive impact on society. The internet can be easily accessed anytime and anywhere, but with the advancement of internet technology, there are many threats lurking in the security of its users. Criminal activities in the digital world are referred to as cybercrime. Numerous cases of cybercrime have occurred worldwide, ranging from attacks that can disable servers to data theft and illegal access. It is noted that more than 50% of companies do not have a plan to respond to these cybercrimes. This is due to various factors, one of which is the limited availability of freely accessible and easily configurable network security platforms for all users. Therefore, this research aims to provide a solution in the form of an open-source-based Intrusion Detection and Prevention System (IDPS) that can be freely distributed and easily configured, one of which is Wazuh. The study uses the Cisco PPDIOO approach in developing a virtual lab with various scenarios for testing and measuring the Quality of Services (QoS) of Wazuh's performance. From the created test scenarios, Wazuh can detect attacks from both inside and outside the network. Wazuh has proven to be capable of detecting and preventing various types of network attacks and features that can facilitate users in responding to cybercrime, making it a potential solution for organizations that have not planned to respond to cybercrime.

A STUDY OF WORLDWIDE PATTERNS IN ALPHABET SIGN LANGUAGE RECOGNITION USING CONVOLUTIONAL AND RECURRENT NEURAL NETWORKS

2025-02-12T11:56:37+00:00

Sign Language Recognition (SLR) has become an essential area of research due to its potential to promote understanding between the deaf and hearing communities through communication. This paper provides an in-depth study of various methodologies and models employed in SLR, focusing on Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). We analyze their application to datasets from various sign languages, such as Arabic Sign Language (ArSL), American Sign Language (ASL), and British Sign Language (BSL), and explore how these models improve the recognition of dynamic, multi-dimensional hand gestures. This research not only advances the understanding of deep learning applications in sign language recognition but also addresses critical challenges in data processing and real-time applications, paving the way for inclusive technologies in informatics and human-computer interaction. Despite the progress in applying deep learning techniques to SLR, several challenges remain, particularly in dataset limitations, handling large vocabularies, and ensuring consistent performance across diverse environments and signers. The paper also investigates the broader applications of SLR, such as virtual reality, healthcare, education, and accessibility, and discusses the integration of SLR with human-computer interaction systems. Furthermore, it highlights current limitations in the field, such as difficulties with video data handling, the need for standard datasets, and issues related to training computational models. Finally, the paper outlines future research directions, including developing more robust SLR systems that can function effectively in uncontrolled environments, improving data collection methodologies, and creating real-time, user-friendly applications to assist the community of deaf and hard-of-hearing individuals.

CLASSIFICATION OF RUPIAH CURRENCY IN THE FORM OF PAPER USING THE MOBILENETV3 LARGE METHOD

2025-02-12T12:08:24+00:00

Money plays an important role in everyday life as a legal tender and a symbol of a country's economic strength. The ability to accurately classify rupiah banknotes has many practical applications such as in automated payment systems, currency exchange, and cash management. However, conventional classification approaches based on digital image processing and image processing techniques are often limited in terms of accuracy and computational efficiency, especially when dealing with a variety of banknote conditions such as wrinkles, stains, or damage. This research aims to propose a new approach by utilising the MobileNetV3 Large architecture, an efficient and lightweight deep learning model, to address the challenges of paper currency classification. The main objective is to improve classification accuracy while minimising computational resources. The dataset used consists of 2873 images of paper rupiah currency of various denominations and conditions from seven classes. These images were processed and trained using the MobileNetV3 Large model that has been customised for this classification task by applying various data augmentation techniques. Experimental results show that the proposed approach is able to achieve 100% classification accuracy on a test dataset with a relatively small model size so that it can be run efficiently on mobile devices or embedded systems. This research makes an important contribution to the development of accurate and efficient rupiah banknote classification techniques for various practical applications in the future.

IMPLEMENTATION OF CONVOLUTIONAL NEURAL NETWORK METHOD IN CLASSIFYING PANDAWA SHADOW PUPPETS

2025-02-12T12:14:59+00:00

The rapid development of technology can lead to the neglect of traditional cultural and artistic aspects by humans. Nonetheless, technology has become integral in society's life. While technology facilitates humans in completing tasks, Negative impacts can also arise. One example of traditional art in Indonesia is shadow puppetry, often featuring stories of the Pandavas from the Mahabharata in puppetry performances. Characters in shadow puppetry are grouped based on character, era, and story, with similar shapes and contours. The similarity of these characters makes them difficult to distinguish and remember. Therefore, an application has been developed that can detect and classify Pandawa shadow puppet characters. The method used in this research is the Convolutional Neural Network (CNN), an effective method in deep learning for classifying data based on informational context The hope is that this application will not only introduce Indonesian culture through Pandawa shadow puppet characters but also provide a high level of accuracy in its classification results. Through the conducted training procedure, the developed model showed an accuracy rate of 95.70%. Furthermore, result verification through the use of a confusion matrix confirmed an accuracy level reaching 88%.

LINE PATH DETECTION ON HIGHWAYS USING THE HOUGH TRANSFORM METHOD

2025-02-12T12:40:20+00:00

Lane line detection on highways is an important problem in the development of intelligent transportation technology or autonomous vehicles. One commonly used method is the Hough Transform method, which is known for its excellent level of accuracy and effectiveness. Line lane detection aims to identify and monitor line lanes on highways, which helps direct and limit vehicle traffic and ensures the safety and efficiency of vehicle movement. This research uses video images from cellphone cameras that have been taken previously. The image is then processed using the Hough Transform algorithm to detect line paths on the highway. The aim of this research is to create a line lane detection system on highways that is able to identify line lanes in various road conditions by utilizing the Hough Transform Algorithm. Apart from that, it also aims to test the ability of the Hough Transform algorithm in the lane line detection system which can provide a warning if the driver is too close to the line lane, increasing safety on the road. Even though there are several obstacles such as poor road conditions, unclear or faded line paths, and busy traffic situations, the results of this research show that the Hough Transform method can be used to detect line paths on highways well, and the level of accuracy is sufficient high namely 83%.

SENTIMENT ANALYSIS OF ONLINE DATING APPS USING SUPPORT VECTOR MACHINE AND NAÏVE BAYES ALGORITHMS

2025-02-12T12:44:33+00:00

In daily life, the use of digital applications is increasingly widespread, making dating apps increasingly popular and an important part of modern social interaction. This research aims to analyze user sentiment towards online dating apps, specifically Tinder, using Support Vector Machine (SVM) and Naïve Bayes algorithms. The problem underlying the importance of this research is the lack of balance between positive and negative sentiments in Tinder app users, which can affect user experience and the quality of service provided by Tinder. Utilizing the CRISP-DM framework, this research involves six stages, from data collection to evaluation. The results showed a significant imbalance between the number of positive and negative sentiments before optimization, but after the application of the SMOTE technique, there was a balancing between the two sentiment categories. SVM achieved 85% accuracy, while Naïve Bayes achieved 84%, with similar performance in identifying positive and negative sentiments. While both models performed satisfactorily, SVM appeared more stable in recognizing both positive and negative sentiments, suggesting the potential to be a superior choice in the context of dating apps. As such, this research makes an important contribution to the understanding of users' views on Timder apps and provides a basis for further development.

USER EXPERIENCE IN METAVERSE BUILDING TRAINING USING PHOENIX-FIRESTORM SOFTWARE

2025-02-12T12:57:59+00:00

This study aims to evaluate the effectiveness of training using Phoenix-Firestorm software in a 3D virtual environment (metaverse) for teachers, lecturers, and students. A total of 49 participants were involved in the online training consisting of seven sessions, facilitated through the Discord platform for voice communication. Each participant was given a virtual area of 35x35 meters for practice, with daily guidance via Discord chat. The training was designed to equip participants with basic skills in building 3D objects, including an understanding of the software and building techniques. After the training, a survey was conducted using a Likert scale of 1-9 to assess participants' understanding of navigation, software customization, virtual communication, and problem-solving. The survey results showed that the majority of participants found Phoenix-Firestorm relatively easy to use, although some challenges were reported regarding the complexity of the interface. These findings will be used as a basis for developing more effective and user-friendly training guidelines in the future, with a focus on improving accessibility and user experience in the context of technology-based learning. This study is in line with previous studies that show the potential of virtual worlds in education, as discussed by Jusuf (2023). Additionally, the use of virtual technology in education is also supported by research on the effectiveness of virtual learning environments, as explained by Wang et al (2022), that digital games contributed to a moderate overall effect size when compared with other instructional methods. These findings are expected to make a significant contribution to the development of innovative training methods in education in the digital era.

COMPARATIVE STUDY OF CNN-BASED ARCHITECTURES ON EYE DISEASES CLASSIFICATION USING FUNDUS IMAGES TO AID OPHTHALMOLOGIST

2025-02-12T12:55:53+00:00

Eye health has a significant impact on quality of life, with more than 2.2 billion people experiencing vision problems. Many of these cases can be prevented or treated. The use of AI for eye disease classification helps healthcare professionals provide optimal care. However, the complexity of fundus images challenges classification performance. This study examines various Convolutional Neural Network (CNN) architectures using Transfer Learning and Adam optimization. Fundus images are processed using CLAHE (clip limit and grid size) and the Wiener filter (size) to enhance contrast and reduce noise. Afterward, ResNet-152, EfficientNet, MobileNetV1, and DenseNet-121 are tested to identify the most effective model. The study aims to determine the optimal CNN architecture for eye disease classification, assisting ophthalmologists in diagnosing eye diseases through fundus images. The best CNN model, ResNet-152, achieved an accuracy of 94.82%, outperforming other models by 3.95 - 8.29%.

LEVERAGING DEEP LEARNING APPROACH FOR ACCURATE ALPHABET RECOGNITION THROUGH HAND GESTURES IN SIGN LANGUAGE

2025-02-12T13:02:32+00:00

Sign language is one way of communication used by people who cannot speak or hear (deaf and speech impaired), so not everyone can understand sign language. Therefore, to facilitate communication between normal people and deaf and speech-impaired people, many systems have been created to translate gestures and signs in sign language into understandable words. Artificial intelligence and computer vision-based technologies, such as YOLOv9 offer solutions to recognize hand gestures more quickly, accurately, and efficiently. This research aims to develop a hand gesture detection system for alphabetic sign language using YOLOv9 architecture, with the aim of improving the accuracy and speed of hand gesture detection. The data used consists of 6500 sign language alphabet hand gesture images that have been labeled with bounding boxes and processed using image augmentation techniques. The model was trained on the Kaggle platform and evaluated using performance metrics such as Accuracy, Precision, Recall, F1-Score, and Intersection over Union (IoU). The results show that the YOLOv9 model achieves an average detection accuracy of 97%, with precision and recall above 90% for most classes. In addition, YOLOv9 shows advantages over other algorithms such as SSD MobileNet v2 and Faster RCNN, both in terms of speed and accuracy. In conclusion, YOLOv9 proved to be very effective in detecting sign language hand gestures, thereby speeding up and facilitating communication. This research is expected to contribute to the development of more inclusive technologies in various fields, such as education, public services, and employment opportunities, which support better communication between sign language users and the general public.

PERFORMANCE EVALUATION OF YOLOV8 IN REAL-TIME VEHICLE DETECTION IN VARIOUS ENVIRONMENTAL CONDITIONS

2025-02-12T13:06:25+00:00

This research focuses on assessing and developing a real-time detection system using the YOLOv8 algorithm. Accurate and fast vehicle detection is a big challenge in modern traffic management, especially in various environmental conditions such as bad weather, low lighting, and high traffic density. The aim of this study was to evaluate the performance of YOLOv8 under these conditions and identify potential improvements. The dataset used consists of 16,990 vehicle images with various variations and environmental conditions. After being trained, the model is evaluated using metrics such as precision, recall, and F1-score, as well as Intersection over Union (IoU) with a threshold of 0.8 on IoU. The results show that YOLOv8 is superior with a fairly high detection accuracy of 78%, with precision of 82% and recall above 90%, and is able to detect vehicles in real-time conditions. However, the challenge of detecting small objects or irregularly shaped vehicles such as tractors still needs to be optimized. This research also compared the performance of YOLOv8 with the SSD (Single Shot Detector) algorithm, where YOLOv8 was proven to be superior in terms of accuracy, precision, recall and F1-score. The research results obtained provide valuable insights for the development of traffic management systems based on deep learning technology. The main contribution of this research is to provide a more efficient and effective vehicle detection solution, which can be applied in modern traffic management systems. Thus, it is hoped that the results of this research can increase the efficiency of traffic management and have a positive impact on the development of intelligent transportation systems in the future.

COMPARISON OF ACCURACY OF VARIOUS TEXT CLASSIFICATION METHODS IN SENTIMENT ANALYSIS OF E-STAMPS AT X

2025-02-12T13:09:28+00:00

In the rapidly evolving digital era, technological innovations are applied in various fields, including law and administration, to improve the efficiency and effectiveness of processes. One of the latest innovations in Indonesia is the implementation of e-metals, which is designed to facilitate legal and secure electronic transactions, and meet the needs of a society that is increasingly dependent on digital technology. Although e-stamps aim to improve efficiency and security in transactions, there are still various perceptions from the public that reflect their views and experiences regarding the implementation of this technology. In this case, sentiment analysis is an effective method to evaluate public opinion generated from text data, such as user reviews and comments on social media. This research aims to analyze the sentiment towards e-metallocations in X app, using text classification methods to separate positive and negative sentiments. After collecting 3282 datasets and performing preprocessing that includes case folding, data cleaning, tokenizing, and stemming, the evaluation results show that the Naive Bayes (GNB) model achieves 96.65% accuracy on training data and 95.28% on testing data. On the other hand, the Support Vector Machine (SVM) model recorded an accuracy of 98.32% on training data and 96.80% on testing data. Meanwhile, the Random Forest model showed a perfect accuracy of 100% on training data and 99.09% on testing data, making it the highest performing model among the three methods tested.

YOLOv9 – BASED TRAFFIC SIGN DETECTION UNDER VARYING LIGHTING CONDITIONS

2025-02-12T13:14:40+00:00

Traffic signs are an important element that functions as a guide, regulator and safety supervisor for road users. In Indonesia, there are various types of traffic signs, including recommendation, prohibition, warning, command, and direction signs, which use numbers, letters, symbols, or a combination of the three to convey clear information to drivers. Based on data from the Indonesian National Police, 148,575 cases of traffic accidents were recorded in 2023, which continues to increase every day due to human error, poor road conditions, and lack of clarity and completeness of signs. This research aims to develop traffic sign detection technology using the YOLOv9 algorithm, starting with collecting 7,980 images from the Roboflow platform, which are then labeled and trained, and evaluated using metrics such as Accuracy, Precision, Recall, F1-Score, and Intersection over Union (IoU ). Then the model was tested to detect traffic signs in various media, such as images and videos. The results of this research show that the YOLO v9 model has the best performance compared to SSD MobileNet v2 and Faster RCNN. The YOLOv9 model achieved an accuracy of 94%, while SSD MobileNet v2 only had an accuracy of 43%, and Faster RCNN had an accuracy of 57%. From the research, it can be concluded that the YOLOv9 model is optimal enough to detect traffic signs in various lighting conditions, because the model has the best performance compared to the other two models, especially in terms of accuracy and balance between precision and recall. This research is expected to support the development of safer autonomous vehicles and intelligent transportation systems through optimal traffic sign detection.

SENTIMENT ANALYSIS OF POST-COVID ONLINE EDUCATION AMONG GEN Z WITH VARIOUS CLASSIFICATION METHODS

2025-02-12T13:18:16+00:00

The COVID-19 pandemic has significantly changed the education sector, shifting from traditional learning to online learning. Generation Z's perception of online education is influenced by their experience as “Digital Natives” who have been familiar with technology since childhood. However, this sudden transition brings new challenges, such as screen fatigue, lack of social interaction, and difficulty in maintaining learning motivation. Sentiment analysis is an important tool to evaluate their experiences and views on online learning. This study aims to investigate Generation Z's views on online education after the pandemic, utilizing various classification methods. Data was collected from Twitter through scraping technique with specific keywords, resulting in a total of 4,986 data obtained using the Tweet Harvest library in Python programming language. The dataset then went through a preprocessing stage, including data cleaning, case folding, tokenizing, stopword removal, and stemming. Before applying Random Forest, SVM, and Naïve Bayes methods, the data is divided into two parts, namely, 3988 training data and 998 testing data with a ratio of 80:20. The accuracy results show that Naïve Bayes achieved 95.49% on training data and 76.05% on testing data, SVM recorded 94.77% accuracy on training data and 87.33% on testing data, and Random Forest obtained 99.97% accuracy on training data and 92.21% on testing data. This research provides important insights into Generation Z's perceptions of post-COVID-19 online education and learning platforms to improve the effectiveness of online learning and identify student challenges in the digital era.

A TOPIC-BASED APPROACH FOR RECOMMENDING UNDERGRADUATE THESIS SUPERVISOR USING LDA WITH COSINE SIMILARITY

2025-02-12T13:25:40+00:00

The thesis is one of the critical factors in determining student graduation. While working on the thesis, students will be guided by a lecturer who has the role and responsibility to ensure that students can prepare the thesis well so that the thesis is ready to be tested and is of good quality. Therefore, selecting a supervisor with the same expertise as the thesis topic is essential in determining students' success in completing their thesis. So far, the selection of thesis supervisors at Dian Nuswantoro University still needs to be done manually by students, so the lack of information about the supervisor can hinder students in determining the supervisor. This study aims to model the topic of lecturer research publications taken from the ResearchGate and Google Scholar platforms so that it is easier for students to choose a thesis supervisor whose research topic is relevant to the student's thesis using the Latent Dirichlet Allocation method. The LDA method will mark each word in the topic in a semi-random distribution. It will calculate the probability of the topic in the dataset and the likelihood of the word against the topic for each iteration. The results of LDA modeling present six main topics of lecturer research with the highest coherence score of 0.764, and then the resulting topics and thesis titles will be compared using cosine similarity. Students can use The highest cosine value as a reference when determining the right thesis topic. Thus, the supervisor selection process will be more focused and in accordance with the student's research interests.

ENHANCING SENTIMENT ANALYSIS OF THE 2024 INDONESIAN PRESIDENTIAL INAUGURATION ON X USING SMOTE-OPTIMIZED NAIVE BAYES CLASSIFIER

2025-02-12T13:30:20+00:00

The inauguration of the President and Vice President of Indonesia for the 2024-2029 period has drawn significant public attention, reflecting widespread political and societal interest. This study aims to optimize sentiment analysis of public opinion on X (formerly Twitter) regarding the inauguration by enhancing the Naïve Bayes Classifier (NBC) with the Synthetic Minority Over-sampling Technique (SMOTE). Addressing the issue of class imbalance in sentiment data, the research demonstrates how SMOTE improves classification robustness. The methodology includes data crawling from X, preprocessing involving tokenization, stemming, and TF-IDF feature extraction, and sentiment labeling using TextBlob. Sentiment classification is conducted with NBC, evaluated under conditions with and without SMOTE. Metrics such as accuracy, precision, recall, and F1-score are utilized to assess performance. Results indicate that the application of SMOTE increases the accuracy of NBC from 98% to 99%, with precision improving from 0.98 to 1 and recall maintaining high levels (0.99). This 1% accuracy enhancement underscores the significance of addressing class imbalance for reliable sentiment analysis. The findings contribute to a better understanding of public sentiment during critical political events and highlight the effectiveness of SMOTE in improving text classification tasks. This research provides valuable insights into leveraging machine learning techniques for analyzing imbalanced datasets, offering implications for both academic and practical applications in sentiment analysis and political studies.

APPLICATION OF VGG16 ARCHITECTURE IN WOOD TYPE CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORK

2025-02-12T13:54:30+00:00

Wood is an important natural resource in construction and the furniture industry, with various types possessing unique characteristics. The selection of wood types is often done manually, which is prone to errors that can negatively impact the working process, product quality, and the sustainability of the forests that source the wood. Therefore, this research aims to improve classification accuracy through the application of technology. This study utilizes Convolutional Neural Network (CNN) with the VGG16 architecture to process images in analyzing the visual characteristics of wood, with the goal of building a model capable of classifying wood types based on images. The dataset used consists of 1,584 samples of wood images sourced from Kaggle. Four models were tested with variations in the training and validation data splits, as well as the use of Adam and Adamax optimizers, over 100 epochs. Model 1 achieved a training accuracy of 96.68% and a testing accuracy of 98.10%. Model 2, with a training accuracy of 99.47% and a testing accuracy of 98.41%, showed the best performance. Models 3 and 4 also yielded testing accuracies of 97.46% and 97.78%, respectively. The results of this study indicate that the application of CNN with the VGG16 architecture can enhance the effectiveness of wood type classification and contribute to more accurate and efficient wood selection practices.

HYBRID METHOD USING NON-NEGATIVE MATRIX FACTORIZATION AND KEYWORD-BASED FILTERING FOR RECOMMENDER SYSTEM IN MOOCS

2025-02-12T14:54:26+00:00

Massive Open Online Courses (MOOCs), introduced by Dave Cormier in 2008, have revolutionized education by providing widespread access to open and participatory online learning. While MOOCs offer broad access and flexibility in learning, users often encounter challenges in selecting appropriate courses. This leads to high dropout rates. To address this issue, this research develops a recommendation system employing the Weighted Hybrid method that combines Non-Negative Matrix Factorization (NMF) and Keyword-Based Filtering (KBF). The primary objective of the research is to enhance the accuracy of course recommendations on MOOCs. The findings of this study demonstrate that the Weighted Hybrid method, integrating NMF and KBF, successfully attained a Mean Average Precision (MAP) of 0.1963. This figure signifies an improvement compared to the MAP value of 0.1855 achieved in prior research. This method effectively addresses challenges such as cold start and sparsity, while also improving scalability. Consequently, the Weighted Hybrid approach holds promise for improving the quality of recommendations, enhancing the user's learning experience, and potentially reducing dropout rates in MOOCs.

IMPLEMENTATION OF THE RANDOM FOREST METHOD FOR CLASSIFYING LUNG X-RAY IMAGE ABNORMALITIES

2025-02-13T01:20:37+00:00

The Covid-19 pandemic has caused a severe global health crisis. Rapid and accurate diagnostics are essential in combating this disease. In this regard, lung X-ray images have become critical for identifying Covid-19 infections. The method used in this study is random forest, a classification method based on ensemble modeling of decision trees. The lung X-ray images used in this study were taken from a datasheet containing images from COVID-19 patients and images from non-Covid-19 patients. The data pre-processing process involves extracting features from the images using image processing techniques and statistical analysis. The random forest model is trained using the processed datasheet to classify the lung X-ray images. The model's performance is evaluated using accuracy, sensitivity, and specificity metrics. In addition, cross-validation is used to measure the reliability and generalization of the model. The study results showed that the random forest method achieved good classification performance in distinguishing COVID-19 lung X-ray images from normal ones. The resulting model provided high accuracy and good sensitivity in identifying Covid-19 cases. These results show the potential of the random forest method in supporting early diagnosis and treatment of COVID-19 disease.

IMPLEMENTATION OF TEXT MINING ON SONG LYRICS FOR SONG CLASSIFICATION BASED ON EMOTION USING WEBSITE-BASED LOGISTIC REGRESSION

2025-02-15T00:13:01+00:00

Music has become an essential medium for expressing emotions and enriching human social experiences. However, the manual interpretation of emotions in song lyrics is often inaccurate and time-consuming, especially for complex or ambiguous lyrics. This creates a need for an automated system that can improve the accuracy and efficiency of emotion classification in song lyrics. Various algorithms, such as K-Nearest Neighbor (K-NN), Naive Bayes Classifier, and Support Vector Machine (SVM), have been applied for emotion classification in song lyrics. Previous research has shown that SVM combined with Particle Swarm Optimization (PSO) achieves an accuracy of up to 90%, while K-NN with feature selection produces the highest f-measure of 66.93%, and Naive Bayes achieves an accuracy of up to 45%. In this study, the Logistic Regression algorithm, supported by the Term Frequency-Inverse Document Frequency (TF-IDF) method, is applied to enhance the accuracy of emotion classification. Evaluation results indicate that the model with figurative language transformation achieves a higher accuracy (93.52%) compared to the model without figurative language transformation (92.31%), demonstrating that figurative language contributes to the richness of emotional expression recognized by the model. This model shows competitive results and can be compared to SVM using PSO while providing better performance than K-NN and Naive Bayes. The system implementation is web-based using the Streamlit framework, allowing users to input lyrics and obtain interactive emotion predictions. This research contributes to the analysis of music emotions and offers an efficient and more accessible alternative for emotion classification in song lyrics.

ENHANCED SPEED AND ACCURACY IN COCOA FRUIT DISEASE IDENTIFICATION USING THE INCEPTION-RESNET CONVOLUTIONAL NEURAL NETWORK (CNN) ALGORITHM

2025-02-20T11:13:07+00:00

The increase in world cocoa consumption is not accompanied by an increase in production, causing a problem of supply shortages in the world. One of the causes of the stagnation in the increase in cocoa production is due to diseases that attack cocoa fruit. The disease can cause unproductive plants, unusable cocoa fruit, and even cause the spread of epidemics in a cocoa fruit garden. One of the preventions that can be done is to identify diseases in cocoa fruit in order to reduce the spread of the disease. The identification process is usually carried out independently by farmers. Identification of cocoa fruit diseases requires knowledge and experience by farmers, so it can cause misidentification or failure to identify the disease. In addition, other factors can arise such as the number of farmers who check, the area of the cocoa fruit garden, and the urgency of identification. To help overcome these problems, a Convolutional Neural Network (CNN) model was developed with the Inception and ResNet architectures. The data used were images obtained from Davao City, Philippines. The model obtained from the analyzed dataset got the best results of 0.99, a specificity value of 0.99, and an F1-score value of 0.99. The model configuration used is a learning rate value of 0.0001, RMSProp optimization function, initialization function (x) He uniform, initialization function (y) Glorot normal, and a batch size of 32.

IMPROVING SHOPPING EXPERIENCES AT NTB MALL THROUGH PERSONALIZED PRODUCT RECOMMENDATIONS USING CONTENT-BASED FILTERING

2025-02-21T06:49:23+00:00

NTB MALL, an e-commerce platform specializing in unique products from micro, small, and medium enterprises (MSMEs) in West Nusa Tenggara, faces challenges in providing personalized product recommendations due to the diversity of its product categories and consumer preferences. To address this, this study implements a content-based filtering (CBF) approach utilizing Term Frequency-Inverse Document Frequency (TF-IDF) and cosine similarity to enhance recommendation accuracy. The system analyzes product attributes and user interaction history to generate tailored suggestions. Experimental results indicate that cosine similarity outperforms Euclidean distance in recommendation precision, achieving an accuracy of 89% and a Mean Reciprocal Rank (MRR) of 95%. Furthermore, user feedback reveals that 93% of users found the recommendations highly relevant, 89% reported increased engagement, and 96% expressed satisfaction with the personalized shopping experience. This research provides a novel application of AI-driven recommendation systems in regional e-commerce marketplaces, demonstrating their potential to improve user experience and foster stronger connections between consumers and local producers.

the ENHANCE OBJECT TRACKING ON AUGMENTED REALITY USING HYBRID CONVOLUTIONAL NEURAL NETWORK AND FAST CORNER DETECTION

2025-02-20T11:15:49+00:00

Markerless augmented reality (AR) is utilized in applications that do not require anchoring to the real world and do not require the use of physical markers (fiducial markers). Augmented object displays not only float but also allow for the automatic placement of 3D augmented reality objects on flat surfaces to enhance realism in real time. There are two challenges that need to be addressed in Markerless AR systems: object tracking and registration, as well as the influence of light intensity. Therefore, the objective of this research is to propose the use of Convolutional Neural Networks (CNN) and Features from Accelerated Segment Test (FAST) corner detection for tracking or detecting objects in markerless augmented reality systems. Testing was conducted using three epoch schemes: 10, 50, and 100. The test results were measured using several parameters, including the execution time, testing loss, and testing accuracy. The test results indicated an improvement in the performance of the tested object detection. The accuracy testing results of using the CNN and FAST corner detection methods were superior to those of the CNN-only method and FAST corner detection alone, reaching 98%. However, this method increases the processing time for object detection. Thus, the processing time of the CNN without FAST corner detection was faster.

A COMPARATIVE STUDY OF MULTI-MASTER REPLICATION OF NOSQL DATABASE SERVER WITH VARYING DATA FORMATS

2025-02-20T11:16:52+00:00

NoSQL Databases are currently an effective solution for managing large data sets distributed across many Servers. NoSQL Database design is usually based on its usability. Specifically related to the system or application to be built. This research aims to measure the Transfer Rate, CPU usage, Memory usage, query execution time for Create, Insert, Delete and remote replication query bandwidth in the Multi-Master Server replication process using two document stored NoSQL Database applications namely CouchBase and CouchDB by entering three different data models namely JSON, XML and CSV. The experimental results show that the Transfer Rate with CSV data format on CouchBase has the lowest value with an average of 111.41 kbps. CPU usage with XML data format on CouchBase has the lowest value with an average of 13.89%. Memory usage with JSON data format on CouchBase has the lowest value with an average of 1.68%. Query Execution Time Create with XML data format on CouchBase has the lowest value with an average of 1.16 seconds. Query Execution Time Insert on CouchBase with CSV data format has the lowest value with an average of 33.28 seconds. Bandwidth Query Execution Time Insert with CSV data format on CouchBase has the lowest value with an average of 24.78 mb. Query Execution Time Delete with JSON, XML and CSV data formats on CouchDB has the lowest value with an average of 1.5 seconds. Further research recommendations are to test Multi-Master Server Replication using other data formats and parameters or test the performance of data migration to other Databases with different data formats.

K-ALLY BASED DYNAMIC FUZZY CLUSTERING FOR GEOPOLITICAL ALLIANCE ANALYSIS: A CASE STUDY INSPIRED BY THE RUSSIAN-UKRAINIAN CONFLICT

2025-03-06T02:24:46+00:00

Geopolitical alliances are often based on a combination of factors such as geographic proximity, military strength, and strategic interests. In this research, we introduce the K-Ally algorithm based on Dynamic Fuzzy Clustering to dynamically analyze alliance patterns between countries. Using fuzzy logic and adaptive thresholds, this algorithm evaluates the potential benefits of alliances based on key attributes, such as geographic distance and power differences. This study is inspired by the allied dynamics that emerged in the Russian-Ukrainian war, where changes in strategy and international relations were key to the continuation of the conflict. The paper also compare this algorithm with the K-Means method commonly used in geopolitical data analysis. Experimental results show that K-Ally based on Dynamic Fuzzy Clustering is able to capture alliance dynamics better than K-Means, especially in conditions of uncertainty or attribute imbalance between countries. This research contributes to the development of new analytical tools for the study of geopolitics and international conflict.

AN EVALUATION OF THE SUCCESSFUL IMPLEMENTATION OF THE INFORMATION SYSTEM PLATFORM MERDEKA MENGAJAR USING HUMAN ORGANIZATION TECHNOLOGY FIT MODEL APPROACH

2025-02-20T11:18:29+00:00

The implementation of technology in education has great potential to improve the quality of learning that supports the implementation of the Merdeka curriculum. The Merdeka Mengajar platform (MMP) is designed to help educators by providing various features including self-development, inspiration and teaching. Uneven ICT infrastructure and teachers' personal abilities are problems in the implementation of the MMP, so it is necessary to analyze the success of the implementation of the MMP. The purpose of this study is to analyze the success of the implementation of the information system for the Merdeka Mengajar Platform by adopting the Hot Fit Model by expanding the Technology component with the ICT Infrastructure variable, expanding the Human component with the personal competence variable, expanding the organizational component with the organizational culture variable and the training & learning variable which can affect the successful implementation of the MMP. The data obtained were 328 respondents who were analyzed using SmartPLS 3.2.9. The analysis results obtained the proposed conceptual model has an accuracy of 58.6%. Net benefits are influenced by system use, user satisfaction, personal competence, structure, environment, organizational culture, and training & learning. Service quality, system quality, information quality, and ICT infrastructure have a positive impact on system use and user satisfaction.

IMPLEMENTATION OF CONVOLUTIONAL NEURAL NETWORK USING MOBILENETV2 TO DISTINGUISH HUMAN AND ARTIFICIAL INTELLIGENCE PAINTING

2025-02-24T08:16:39+00:00

The advancement of artificial intelligence technology has had a significant impact on various fields, including painting. Artificial intelligence is now able to create works of art that resemble paintings produced by humans with a high level of detail and complexity. However, this progress has also created new problems in the world of painting, namely the difficulty in distinguishing between works produced by humans and those created by artificial intelligence. This problem has an impact on the originality of the artwork and has implications for aspects of ethics and creativity. This study aims to develop a deep learning model that can classify human and artificial intelligence paintings, and overcome the challenges in distinguishing between the two. The methodology used is the Cross Industry Standard Process for Data Mining (CRISP-DM), with a dataset consisting of 1,000 painting images. The architecture used is MobileNetV2, implemented using TensorFlow to build a Convolutional Neural Network (CNN). Techniques such as data preparation, data labeling, data splitting, resizing, and data augmentation are applied to improve model performance. Six test scenarios were carried out with variations in the learning rate, number of epochs, and freeze or unfreeze configurations on the base model. The results showed that the best model with a learning rate of 0.0001, base model unfreeze, and 5 epochs managed to achieve an accuracy of 97%, without any indication of overfitting or underfitting. This model was then implemented on an Android application in TFLite format, which can predict image classes with a confidence level of 89.98%.

COMPARATIVE ANALYSIS OF PERFORMANCE AND EFFICIENCY OF LOAD BALANCING ALGORITHMS ON INGRESS CONTROLLER

2025-03-06T02:54:17+00:00

Kubernetes has become the dominant container orchestration platform in production environments, with the ingress controller playing a critical role in managing external traffic to services within the cluster. This study aims to provide recommendations for optimal load balancing algorithms for Kubernetes production environments by analyzing and comparing the performance of four algorithms namely round robin, static-rr, least connection, and random on the HAProxy ingress controller. The research method is conducted through observation using k6 and Grafana performance test tools, as well as literature studies, with measurements including total requests, throughput, latency, CPU usage, and memory at various levels of user load. The data was analyzed using descriptive statistical techniques, normality test, homogeneity test, and tests for group differences using one-way ANOVA or Kruskal-Wallis H. The results show that static-rr excels in throughput, total requests, and CPU and memory efficiency at high load, while least connection is more effective for latency at low load. Round robin and random showed stable performance at low load but less optimal at high load. The conclusion of this study is that choosing the right load balancing algorithm depends on the load characteristics and desired performance metrics, to ensure optimal Kubernetes performance under various load scenarios in production environments.