Jurnal Teknik Informatika (Jutif)

OPTIMIZATION OF DISEASE PREDICTION ACCURACY THROUGH ARTIFICIAL NEURAL NETWORK (ANN) ALGORITHMS IN DIAGNESE APPLICATION

Ruvita Faurina — 2024-04-03

This research aims to enhance the accuracy and speed of diagnoses in the Diagnese application by implementing the ANN algorithm for disease prediction. The dataset used for experimentation was featuring binary data types, containing 131 symptoms used to predict 41 types of diseases. The Diagnese application assists patients in identifying diseases and finding suitable specialist doctors based on reported symptoms. To achieve this goal, researchers explored various machine learning algorithms, such as decision trees, SVM, Random Forest, Logistic Regression, and ANN. Through comprehensive analysis, the ANN algorithm outperforms other algorithms and showcases the best performance. The research results demonstrate that integrating this application can significantly improve diagnostic accuracy and speed, thereby potentially reducing treatment delays and enhancing patient health outcomes. The neural network model displayed exceptional accuracy across training, validation, and testing datasets, scoring 97%, 99%, and 95%, respectively. Overall, this study showcases the potential of implementing the ANN algorithm within Diagnese applications to elevate the accuracy and efficiency of disease diagnosis. The application of this model is expected to augment the efficiency and precision of the medical diagnosis process, enabling doctors to make more accurate decisions and provide more effective patient care.

PATTERN CLASSIFICATION SIGN LANGUAGE USING FEATURES DESCRIPTORS AND MACHINE LEARNING

Nurhadi Nurhadi — 2024-04-03

Sign language is way of communication for the deaf and speech impaired. In Indonesia, the utilization of a standardized language involves the incorporation of American Sign Language (ASL). ASL is employed for various communication needs, ranging from basic alphanumeric fingerspelling (A-Z and numbers) to the more complex SIBI form (comprising gesture vocabulary) in everyday interactions as well as formal contexts. This surge in the digitization of sign language underscores the ongoing advancements in research and development. The challenge in this research lies in the ability to recognize American Sign Language (ASL) with diverse intensities and invariant backgrounds. Therefore, the study emphasis is on proposing a suitable segmentation method comparison for multi-intensity ASL cases. Subsequently, global feature descriptor methods, including Color Histogram, Hu Moments, and Haralick Texture techniques, are applied for feature extraction. The result of the Logistic Regression method versus the supervised Random Forest checks accuracy and suitability in identifying ASL fingerspelling. The findings of this research is predictive value of logistic regression is 48%, with class Y having the highest precision (0.86), class V having the lowest accuracy (0.16), and class L having the highest recall (0.73). The maximum precision in classes B, F, H, I, K, Y, and Z is 1.00, and the lowest in class U is 0.58, while the highest recall is in class G, which is 1.00. The lowest is in class V, while the predictive value from the random forest is 86 percent. Class H has the greatest f1 score (0.99), while class U has the lowest f1 score (0.64). The Random Forest method outperforms the two methods suggested in the paper, according to the comparison.

DOCKER-BASED MONOLITHIC AND MICROSERVICES ARCHITECTURE PERFORMANCE COMPARISON

Deni Panji Dirgantara — 2024-04-04

Most developers still use the monolithic architecture, where all components of an application are combined into one integrated system, so each part depends on other components. The monolithic architecture has weaknesses, such as when a failure occurs in one component, all parts cannot be executed because each component relies on one other component. Microservices can be a solution to this, considering that in the microservices architecture, each element or service is created and put separately, so when a failure occurs in one component, other components will not be affected and can still run normally. This research aims to determine the implementation and performance comparison between monolithic architecture and microservices Architecture in the Agreeculture Market web app. Agreeculture Market is a web application that aims to facilitate the transaction process of agricultural commodities and make it easier for agricultural commodity producers to market their products. The measurement method used to measure the performance of both architectures is load testing using JMeter and performance tools from task manager and comparing the response time, throughput, disk usage, CPU usage, and memory usage of both used architectures. With two measurement schemes with Docker and without Docker, the result of this research is a performance comparison between the two architectures, where the backend application Agreeculture Market, which uses microservices architecture with Docker and API gateway, performs better than the monolithic architecture version. Conversely, the monolithic architecture performs better than the microservices architecture in the scheme without Docker and API gateway.

OPTIMIZING RAW MATERIAL INVENTORY MANAGEMENT OF MSME PRODUCT USING EXTREME GRADIENT BOOSTING (XGBOOST) REGRESSOR ALGORITHM: A SALES PREDICTION APPROACH

Muhammad Khusni Fikri — 2024-04-04

Micro, Small and Medium Enterprises or MSMEs have a very important role for the survival of the economic sector in Indonesia. However, as the development of MSMEs, followed by a series of problems that arise. One of them is the problem of sales, business people have difficulty in determining the number of product sales in the future so that there is often an accumulation of raw materials or unsold products. This study aims to help MSMEs optimize raw material management by predicting product sales using the XGBoost Regressor Algorithm. Recently, the algorithm is very famous in the competition because of its reliability and no one has applied it to predict MSME product sales. Based on several other studies, this algorithm is accurate in predicting a value, such as predicting stock prices and the number of accidents in Bali, Indonesia. This research uses historical product sales data and weather data consisting of air temperature and relative humidity in Semarang Indonesia to train and evaluate the performance of the model. The prediction model was performed with predetermined variables and resulted in MAE 3.0752730568649156, MSE 38.25842541629838, and RMSE 6.185339555456788. In the end, it is concluded that the model built with XGBoost Regressor has a low error rate so that it can accurately predict the sales of an MSME product and optimize the management of raw materials for related products.

COMMUNICATION SECURITY IN THE MQTT PROTOCOL FOR MONITORING INTERNET OF THINGS DEVICES USING METHODS ELLIPTIC CURVE CRYPTOGRAPHY

Axel Natanael Salim — 2024-04-04

The emergence of the IoT has become one of the most significant technology trends. The application of IoT is aimed at enhancing efficiency, comfort, and facilitating various human activities. One key aspect of IoT implementation is efficient communication between devices, with one of the most commonly used protocols being MQTT protocol. MQTT enables the transmission of data in real-time or based on specific events, although there are still several challenges that need to be addressed. One of the main challenges of MQTT is information security issues, prompting this research to examine effective solutions to enhance communication security in IoT applications that utilize MQTT protocol. One method of securing communication between IoT devices can involve using lightweight cryptographic communication security methods such as ECC method. ECC method is chosen because it utilizes shorter keys while still providing high security, making it more efficient when implemented on IoT devices. The results obtained indicate that data sent to MQTT Broker cannot be read and converted manually, ensuring much safer data transmission. Based on the test results, the tool can effectively read, process, and send data to MQTT Broker. QoS measurements on the system revealed that data encrypted and sent from the subscriber to MQTT Broker had an average delay time of 54.1 ms, throughput of 410.4 bps, zero packet loss, and jitter of 0.00 ms. Looking at the research findings, it can be concluded that this ECC method could serve as a solution to data communication security issues in the MQTT protocol.

GRAPHICAL COMPUTING FOR BATIK PATTERN DESIGN BASED ON L-SYSTEM

Eka Wahyu Hidayat — 2024-04-04

The challenge faced by the batik industry in the industrial era 5.0 is the adaptation to technology in the production process. One way to overcome this challenge is to start from the basics in the batik industry, namely the creative process of designing batik patterns. It is important to pay special attention to this process to enhance digital transformation in the batik industry. The purpose of this paper is to present the design and creation of batik patterns using the L-System-based fractal approach. Previous research has shown that the L-System can be used to model plant growth in 2D and 3D contexts. In a similar way, the L-System is used in this study to create batik patterns. Experiments were conducted through three stages, namely Data Acquisition, Data Identification, and Modeling. The experiment results in a dataset of batik motifs that can be used as parameters to replace line segments in the L-System. The design and creation of batik patterns using the L-System only needs to be done once, so that from one pattern, a variety of different motifs can be produced easily by simply changing the parameters. This shows that the design and creation of batik patterns using L-System is more efficient and practical. In addition, the fractal dimension calculation is used to understand and describe the fractal properties of the resulting objects. In this study, it was found that there are four motifs without ornaments that have higher fractal dimension values than motifs with equivalent ornaments.

A ROBUST AND IMPERCEPTIBLE FOR DIGITAL IMAGE ENCRYPTION USING CHACHA20

Widhi Bagus Nugroho — 2024-04-04

In the current era, data security is mandatory because it protects our personal data from being used by irresponsible people. The objective of this research is to show the robustness of the method we propose to encrypt images using the chacha20 algorithm which is included in the symmetric encryption cryptography technique and uses one key for both encryption and decryption processes. we use the encryption method by reading the bits from a digital image which is processed using the chacha20 algorithm to get the results of the digital image encryption. The results of this study indicate that the Chacha20 algorithm is secure to use when encrypting and decrypting digital images. The average MSE value generated by the chacha20 algorithm is 0.1232. The average PSNR value is 57.4784. The average value of UACI is 49.99%. The average value of NPCR is 99.602%. The test values were acquired by executing encryption and decryption processes on 5 distinct colour digital images with different size. Additionally, this study displays histograms for the original digital image, as well as for the encrypted and decrypted digital images, illustrating the pixel distribution in each. The histogram also serves as material for analysis of the success of the encryption and decryption processes in digital images.

IMAGE DATA SECURITY USING VERNAM CIPHER ALGORITHM

Supiyanto Supiyanto — 2024-04-04

The Vernam Cipher algorithm is a symmetric key algorithm, as it uses the same key for encryption and decryption. It utilizes a binary number system with XOR operation to produce a series of bits. This study aims to implement the Vernam cipher algorithm to secure personal and confidential image data, which is at risk of misuse when shared through chat applications like Facebook, WhatsApp, and email. Therefore, developing image protection applications is crucial. The research explores whether the vernam cipher algorithm, working with single bits in block form and based on binary numbers, can effectively secure image data, specifically grey scale images with BMP and JPG extensions. The approach involves applying the Vernam cipher algorithm to programming language to create a data security application. The outcome is an image security application program, with test results indicating successful encryption with significant randomness. The decryption process with the Vernam cipher method can restore encrypted images to their original state, although some distortion may occur, especially with JPEG images. Decryption of BMP images is nearly flawless. The key for data security can vary in length and form, with encryption taking longer than decryption.

SURICATA ACCURACY OPTIMIZATION BASED ON LIVE ANALYSIS USING ONE-CLASS SUPPORT VECTOR MACHINE METHOD AND STREAMLIT FRAMEWORK

I Putu Yesha Agus Ariwanta — 2024-04-04

Based on data from the Checkpoint website, there are more than 10 million cyber-attacks in a single day, and the top sequence of this cyber-attack is evident in educational institutions. The IT unit of Kartini Bali Health Polytechnic has not yet conducted testing for accuracy and speed to detect suspicious activities on the computer network. The implementation of network security systems that have not undergone testing will undoubtedly have a negative impact on system providers and users. The application of Live Analysis based on a website and the One-Class Support Vector Machine (SVM) is used to optimize the capabilities of the Suricata in detecting suspicious activities on computer networks and providing visual and real-time reports. This research utilizes the Suricata for optimizing the computer network security system, with the researcher using the Streamlit Framework for Live Analysis based on a website and the One-Class Support Vector Machine (SVM) for classifying log data and visual reporting. For testing the computer network security system, tools such as Nmap, Loic, and Brutus are used. The results of the research using the One-Class Support Vector Machine (SVM) in detecting three types of attacks Port Scanning, DDOS Attack, and Brute Force Attack, show an accuracy value of 96%, precision of 95%, recall of 96%, and F1-Score of 95%. In the performance and load testing of the live analysis system using the Streamlit framework, the results show that the developed system is responsive, with CPU usage at 38%, memory usage at 62.3%, and an average system load time of 5 milliseconds.

THALASSEMIA MINOR SCREENING APPLICATION USING THE C4.5 METHOD BASED ON LARAVEL

Nicolas Sohputro — 2024-04-04

Thalassemia is an inherited blood disorder that causes anaemia and weak red blood cells. Thalassemia minor is a type of thalassemia where the patient is a carrier of thalassemia and only experiences mild anaemia. To prevent an increase in the number of thalassemia cases, a screening process is held for an individual to confirm whether there is a thalassemia carrier in the body. In providing screening in Banyumas Regency, the Unsoed Medical Faculty Thalassemia Research Team encountered several problems, namely that the screening results could only show whether an individual was a carrier of thalassemia minor or not. This causes a problem because a good screening result is a probability. The second problem is the absence of an integrated information system for thalassemia control in Banyumas Regency. The solution to these two problems is to build a thalassemia minor screening application. The application uses the C4.5 data mining method to calculate the likelihood of thalassemia minor in individuals. The application is made website-based using Laravel to speed up website development. The system also uses a web service to be able to access the created C4.5 algorithm.

REDUCING UNDER-FETCHING AND OVER-FETCHING IN REST API WITH GRAPHQL FOR WEB-BASED SOFTWARE DEVELOPMENT

Rizki Nuzul Muzaki — 2024-04-04

Rest API is the most popular architectural style in website-based software development. However, Rest API has under-fetching and over-fetching problems. Under-fetching is a situation when the client has to make requests to several endpoints, while over-fetching is a situation when the client receives more data than needed. There is an alternative technology to Rest API, namely GraphQL. GraphQL has the potential to solve both under-fetching and over-fetching problems. This research aims to analyze how quickly GraphQL responds in overcoming under-fetching and over-fetching problems and conducting condition analysis to determine when it is best to use GraphQL. In this research, tests were conducted to answer these problems by applying each of the five test scenarios for under-fetching and over-fetching problems. Test results show that GraphQL can provide response speeds of 36.84% to 93.04% superior to Rest API. In the case of under-fetching, it is best to choose GraphQL when there is a need to call more than four endpoints. Meanwhile, for over-fetching problems, using the Rest API provides adequate response speed. However, if a more optimal response speed is needed, using GraphQL could be an alternative.

PERANALYSIS OF ACADEMIC WEBSITE USING WEBQUAL 4.0 METHOD AND IMPORTANCE-PERFORMANCE ANALYSIS (IPA)

Fizal Okta Andrean — 2024-04-04

The Academic Website plays a crucial role as the primary channel for delivering academic information to the entire academic community. Its main functions include providing vital information such as Graduation Schedules, Academic Year Calendars, and Scholarship Announcements, making it an indispensable source of information for students. The quality of services on this website is a crucial aspect in meeting the information needs of students. This research aims to evaluate and enhance the quality of the website, with a primary focus on improving services for students and achieving a higher ranking in Webometrics State Islamic Religious Higher Education Institution (PTKIN), currently positioned at 18th. The research methodology utilizes WebQual 4.0 to assess the website's quality, focusing on usability, information quality, and service interaction quality. The Importance-Performance Analysis (IPA) approach is employed to guide the website's development based on the importance and actual performance of each quality attribute. The Webqual Index analysis results indicate that the website achieves a score of 0.85 or 85%, highlighting good service quality but also indicating the need for improvement in information and service interaction quality. This study produces a comprehensive guide for the necessary changes and developments in the Academic Website. The guide ensures that the website aligns with the dynamic needs of the university community, creating a virtual environment that supports and facilitates access to information for students. These improvements are expected not only to enhance the Webometrics PTKIN ranking but also to increase student satisfaction and engagement in the academic process

INFORMATION TECHNOLOGY GOVERNANCE ANALYSIS USING COBIT 5 FRAMEWORK AT SMPN 18 BANDAR LAMPUNG

Salsabila Indriyani — 2024-04-04

So far, the management of Information Technology at SMPN 18 Bandar Lampung has not held an information technology governance analysis, so that the application of information technology infrastructure cannot be known at the maturity level. This study aims to determine the level of maturity of the application of information technology required information technology governance analysis. The method used in the COBIT 5 framework is up to phase 4 - Plan Programe, the calculation used is by finding the statistical average or mean value in the form of the total value of the various items contained in the questionnaire. The results of this research the average maturity index value is 3.4 and (maturity level as is) in the APO, BAI, and MEA domains, at level 3 in the APO, BAI, MEA domains. Based on the results of the research, the researcher provides suggestions regarding the procedures chosen based on the research findings to help the information technology infrastructure of SMPN 18 Bandar Lampung reach the required maturity level.

PREDICTION OF 2024 PRESIDENTIAL ELECTION USING K-NN WITH METRIC APPROACHES CHEBYSHEV AND EUCLIDEAN BASED ON TWITTER DATA INVESTIGATION

Steven Ryan Darmawan — 2024-04-04

The potential difference between the popularity of presidential candidates on social media and in the general public poses a serious challenge in predicting the outcome of the 2024 presidential election. Technical constraints in collecting, cleaning and analyzing dynamic and large-scale social media data can threaten the accuracy and validity of predictions. To overcome this problem, careful steps and in-depth understanding are needed. Therefore, this study aims to predict the winner of the 2024 presidential election from the popularity of presidential candidates Anies Baswedan, Ganjar Pranowo, and Prabowo Subianto on Twitter. The K-Nearest Neighbor (K-NN) method with the Both Metric approach (Euclidean and Chebyshev) was used to analyze 51,192 tweet data through the Knowledge Discovery in Database (KDD) stage using Orange software. The evaluation results show almost the same performance, with AUC values of 0.725 for Euclidean and 0.720 for Chebyshev. The CA result was 55.6% for Euclidean and 55.4% for Chebyshev. Although F1, precision, and recall were almost the same, overall, the Euclidean metric was better. The prediction shows Prabowo Subianto as the most popular candidate on Twitter. Nonetheless, these results need to be interpreted with caution and strengthened with further analysis and additional data to get a more comprehensive conclusion. This research shows that K-NN with both metrics can provide predictions above 50%, reliable enough to be able to predict the most popular candidates on Twitter.

IMPLEMENTATION OF DEEP LEARNING ON FLOWER CLASSIFICATION USING CNN METHOD

Anggun Pratiwi — 2024-04-04

Technological developments in the field of artificial intelligence, particularly deep learning, have made significant contributions to various applications, including pattern recognition and object classification in visual images. One of the interesting applications of deep learning is image classification, where these techniques have proven effective in tackling complex problems, such as object recognition in visual images. Flowering plants, with approximately 369,000 known species, are an interesting object of study. In an effort to classify different types of flowers quickly and efficiently, a digital approach is a must. This research aims to implement deep learning technology, especially CNN method, in flower classification. One method that can be used is Convolutional Neural Network (CNN), a deep learning algorithm that is able to process image information well. In flower type classification, supervised learning techniques are essential. By giving the label (flower type) to the algorithm as the basis of truth, the use of CNN on a large scale can produce predictions and classifications with a high level of accuracy. This research produces a classification model that is more precise and able to overcome variations in flower morphology with 2 different datasets namely Oxford17 resulting in 84% accuracy and oxford102 resulting in an accuracy value of 64%.

CUSTOMER LOYALTY SEGMENTATION IN ONLINE STORE USING LRFM AND MLRFM IN COMBINATION WITH RM K-MEANS ALGORITHM

Angelina Caroline Utomo — 2024-04-15

The rapid development of online business in recent years has driven Store X to embark on a digital transformation. By the end of 2020, Store X relocate their conventional business to online business. The greatest obstacle and key to success for online business operators, such as Store X, is gaining and retaining consumer loyalty in the face of an increasing number of competitors. Therefore, the company must be able to identify the character (behavior) of its clients to provide appropriate treatment. Each customer's behavior is unique, which means they must all be treated differently. However, all this time, Online Store X has provided the same treatment (as much of a discount) to all its customers due to the lack of information regarding their customers’ characteristics. Therefore, in this study, customers of Online Store X were segmented based on their transactional behavior using online transaction history data from March 2021 to March 2023. Two customer analysis models, LRFM and MLRFM, will be combined with RM K-Means to find the best combination through Silhouette Coefficient values. The optimal number of clusters (k) is then determined using the Elbow Method. The results indicate that the optimal number of clusters for both combinations is K=3, with the combination of MLRFM and RM K-Means is the best combination. The finest combination has a silhouette coefficient value of 0.8609. Based on this combination, it is also known that 2,053 customers in cluster 3 are loyal customers, while 2,339 customers in cluster 1 and 2 are lost customers. The results of this study were also implemented on websites built for X Store using Python programming languages and MySQL databases, making it easier for companies to see data visualization.

CATTLE BODY WEIGHT PREDICTION USING REGRESSION MACHINE LEARNING

Anjar Setiawan — 2024-04-15

Increasing efficiency and productivity in the cattle farming industry can have a significant economic impact. Cow health and productivity problems directly impact the quality of the meat and milk produced. In the cattle farming industry, it can help predict cow weight oriented to beef and milk quality. The importance of predicting cow weight for farmers is to monitor animal development. Meanwhile, for traders, knowing the animal's weight makes it easier to calculate the price of the animal meat they buy. This research aims to predict cow weight by increasing the results of smaller MAE values. The methods used are linear Regressor (LR), Random Forest Regressor (RFR), Support Vector Regressor (SVR), K-Neighbors Regressor (KNR), Multi-layer Perceptron Regressor (MLPR), Gradient Boosting Regressor (GBR), Light Gradient boosting (LGB), and extreme gradient boosting regressor (XGBR). Producing cattle weight predictions using the SVR method produces the best values, namely mean absolute error (MAE) of 0.09 kg, mean absolute perception error (MAPE) of 0.02%, root mean square error (RMSE) of 0.08 kg, and R-square of 0.97 compared to with other algorithm methods and the results of statistical correlation analysis showed several significant relationships between morphometric variables and live weight.

STREAM CIPHER ALGORITHM FOR ENCRYPTING TEXT USING LOGISTIC MAP, AUTO PARAMETERS LINEAR CONGRUENTIAL GENERATOR (APLCG), AND GRAY CODE

Adriana Fanggidae — 2024-04-15

One aspect frequently posing a challenge in cryptography pertains to the length of the secret key that users must remember. Achieving the requisite key length for cryptographic algorithms necessitates key padding. However, it is crucial to note that key padding is susceptible to predictable patterns. Both the Linear Congruential Generator (LCG) and gray code are algorithms employed to generate sequences of padded key bits. Regrettably, LCG requires the determination of two pre-defined parameters, whereas the Auto Parameters Linear Congruential Generator (APLCG) automatically establishes these parameters. These parameters play a pivotal role in generating unique sequences of random integers. To fortify key security, the generation of new keys is performed using a modified logistic map, an enhancement of the standard logistic map that exhibits random behavior consistently. Stream cipher, an encryption algorithm, necessitates a continuous key stream matching the bit or byte length of the message. We conducted experiments on stream cipher algorithms employing key streams generated from APLCG, gray code, and modified logistic map. Twenty text documents were utilized as test samples. The outcomes indicate that stream ciphers employing APLCG, gray code, and modified logistic map demonstrate high-security performance based on the statistical analysis conducted.

BOARDING HOUSE RECOMMENDATION WITH COLLABORATIVE FILTERING USING THE GENERATIVE ADVERSARIAL NETWORKS (GANS) METHOD

Mohammad Fajra Septariken — 2024-04-15

This research represents a concerted effort to tackle the pressing challenge of facilitating a personalized and efficient boarding house recommendation system tailored to individual user preferences, particularly among students. The overarching objective is to streamline and simplify the often arduous task of locating suitable accommodations by harnessing the potential of Collaborative Filtering. The deliberate selection of Collaborative Filtering as the cornerstone of this recommendation system stems from its proven efficacy in scrutinizing intricate user behavior patterns and deriving precise, tailored recommendations. Leveraging historical boarding house data, this methodology meticulously identifies patterns and similarities among users to offer suggestions finely aligned with their specific preferences. Integral to this research methodology is the concurrent utilization of Generative Adversarial Networks (GANs), serving a pivotal role in evaluating the system's accuracy. This dual-pronged approach, amalgamating Collaborative Filtering for recommendation generation and GANs for accuracy assessment, aims to ensure the system's efficacy in delivering precise, individualized suggestions. The findings of this study underscore a promising outcome – a system proficient in furnishing boarding house recommendations remarkably attuned to user preferences. This system's potential transcends the realm of student housing, presenting opportunities for broader applications across diverse fields requiring personalized recommendation systems. Crucially, the study's meticulous optimization of the GANs model, involving meticulous parameter adjustments including epoch count, optimizer selection (Adam), employment of mean absolute error (MAE) function, and fine-tuning a learning rate of 0.002, culminated in an outstanding achievement. The resultant MAE value of 0.0180 denotes minimal prediction errors, signifying estimations remarkably proximate to actual test data values, thus solidifying the system's reliability and precision. Ultimately, the successful development and evaluation of this boarding house recommendation system hold profound implications, promising to significantly enhance student experiences in discovering accommodations aligned with their preferences. Furthermore, this study's methodological approach paves the way for future research and wider applications in diverse domains seeking effective, personalized recommendation systems.

COMBINATION K-MEANS AND LSTM FOR SOCIAL MEDIA BLACK CAMPAIGN DETECTION OF INDONESIA PRESIDENTIAL CANDIDATES 2024

Wisnu Priambodo — 2024-04-15

Social media has become the main platform for the public and political figures to voice opinions and run political campaigns. Despite its positive impact, social media also has negative impacts, particularly in the spread of Black Campaigns. This phenomenon has become critical, especially about the 2024 elections in Indonesia that target presidential candidates. Black campaigns can trigger conflict and damage the image of presidential candidates in the eyes of the public. Therefore, it is important to detect black campaigns against presidential candidates. This research develops a Black Campaign detection model using the K-means clustering algorithm and the Long Short-Term Memory (LSTM) approach. K-means is implemented to cluster text data on Twitter social media, while LSTM is used to learn word order patterns and detect text. The result is that K-means can effectively prepare the data, and classification using LSTM shows an accuracy of 90.28%. The comparison with Ensemble Learning classification model achieved an accuracy of 94.31%. Evaluation involved accuracy, precision, recall, and F1-score, with the result that Ensemble Learning was slightly superior in the evaluation matrix. However, compared to Ensemble Learning, LSTM has an advantage in understanding word order, which can be achieved by utilizing the advantages of Deep Learning Recurrent Neural Network architecture. Testing on sample data shows the similarity between LSTM and Ensemble Learning models in detecting Black Campaigns on Twitter social media post text data.

LOW CODE INTEGRATION TESTING IN OUTSYSTEMS PERSONAL ENVIRONMENT

I Dewa Ayu Indira Wulandari Chrisna — 2024-04-15

As implied by its name, low code platforms enable software development with minimal or no coding involved. Consequently, ensuring the correctness of the software becomes crucial as developers are unable to directly scrutinize the logic. Furthermore, discussions about the various testing approaches applicable to such applications are relatively scarce. This study aims to conduct integration testing through both white box and black box methods, as well as exploring the types of testing that can be carried out on low code based applications. This research involves several stages, including creating a basic e-shop application and API using OutSystems, test preparation, and test execution. API testing utilizes OutSystems' BDDFramework and Postman automation testing tools, while web page integration is carried out using Katalon Studio. The test results indicate only one of the total 23 test cases was considered failed because the result did not match the expected result. Apart from that, of the four existing levels of testing, component testing can also be carried out on the OutSystems application. However, only with the black box testing method because testing is carried out without accessing the program source code. The comparative execution of API testing (white box) using two distinct testing tools reveals the superior effectiveness of Postman over BDDFramework, offering more comprehensive test outcomes and enhanced test case coverage. In the realm of UI integration testing, Katalon Studio emerges as a fitting tool, benefiting from its record and replay feature that facilitates the definition of test steps.

ANALYSIS OF GRABAG GUIDE APPLICATION ACCEPTANCE FOR INTRODUCTION TO TOURIST ATTRACTIONS USING THE TECHNOLOGY ACCEPTANCE MODEL (TAM)

Yusuf Wahyu Setiya Putra — 2024-04-15

This research aims to analyze application user acceptance of the tourism introduction Android application in Grabag District, Magelang Regency using three (3) variables contained in the TAM (Technology Acceptance Model) model, namely Attitude towards Use, Perception of Ease of Use and Perception of Usefulness. This research needs to be carried out to resolve the problem of application acceptance which has an impact on the level of tourist visits, so that later application development and improvements can be carried out according to needs. The respondents used were 81 users consisting of sub-district officers, business or tourism owners and tourists. To obtain data whose validity and reliability have been tested and then analyzed using multiple linear regression techniques, a data collection method using a questionnaire was used. The results of the analysis of the 1st regression equation show that the Perception of Usefulness variable (X1) has a significant influence on the Attitude towards Use variable (Y). The Perception of User Ease variable (X2) has a significant influence on the Attitude towards Use variable (Y). The results of the analysis of the 2nd regression equation show that variables X1 and X2 together have a significant influence on Attitude to Use (Y).

LEAF DISEASE DETECTION IN TOMATO PLANTS USING XCEPTION MODEL IN CONVOLUTIONAL NEURAL NETWORK METHOD

Nurhikma Arifin — 2024-04-15

This study aims to detect leaf diseases in tomato plants by applying the Xception model in the Convolutional Neural Network (CNN) method. The study categorizes tomato conditions into three main categories: Early Blight, Late Blight, and Healthy. Early Blight is generally infected by specific pathogens that cause spots and damage in the early stages of plant growth, while Late Blight is infected by pathogens in the later stages of the growing season. Meanwhile, the healthy category indicates normal conditions without disease symptoms. The dataset used consists of 300 tomato images, with each category having 100 images. In the model training phase using the fit method in TensorFlow, 17 epochs were performed to teach the model to recognize patterns in tomato leaf disease images in the training dataset. The model testing results on 30 tomato leaf images showed an accuracy rate of 85.84%. This result indicates a positive indication that the developed CNN model performs well in detecting and classifying tomato leaf conditions. Thus, this research can contribute to improving the understanding and management of leaf diseases in tomato plants to support more productive and sustainable agriculture.

ANALYSIS AND IMPLEMENTATION OF YOLOV7 IN DETECTING PIN DEL IN REAL-TIME

Iustisia Natalia Simbolon — 2024-04-15

Real-time object detection is the process of identifying and tracking objects instantly and directly without any delay between image input and output. Carrying out real-time detection is a challenge in detection systems because it requires speed and accuracy of detection. This research proposes the application of the YOLOv7 algorithm which allows object localization and classification in one stage. This detection is carried out in real time on two objects, namely PinDel and Students. This research focuses on applying the YOLOv7 algorithm to detect real-time use of Pin Del by students. In this research, several hyperparameters were adjusted until the optimal value was found, including epoch with a value of 300, as well as confidence threshold, and IoU threshold with a value of 0.5. The model evaluation results from hyperparameter experiments show good results, with precision of 0.946, recall of 0.959, and mAP@0.5 of 0.977. This research has succeeded in detecting Pin Del objects in real time by obtaining a detection speed of between 7 and 40 FPS, which shows a fast response in detecting objects in real time. This research has contributed to the development of real-time object detection technology and its application in Pin Del use cases by students.

RECOMMENDATION SYSTEM TO SELECT A MAJOR OF VOCATIONAL SCHOOL USING DECISION TREE

Ardiansyah Risko Anwari — 2024-04-17

A recommendation system is a tool that can be used to provide suggestions to users about something they are interested in, such as products, content, music, movies, or even majors at school. When registering for majors at vocational high school (SMK), some students sometimes difficult to select major based on their interests and abilities. This study aims to develop a recommendation system to select major in SMK, so that it can help prospective students choose majors according to their abilities. The method used is Research and Development (R&D), using the waterfall development model which consists of several stages, namely requirements analysis, system design, design implementation, and system testing. The algorithm used to recommend choices is a decision tree, a predictive model that maps input data to output targets based on a series of decisions or separation rules. The parameters used to recommend the selection of majors are the value data of last year's applicants. The evaluation was carried out using the system usability scale (SUS) involving 25 participants (17 males and 8 females) aged from 14 to 16 years old. Based on the analysis carried out, the results showed that SUS score is 89.7, which means that included to the excellent category in measuring adjective ranges, and acceptable in the acceptability scale. Thus it can be concluded that this department recommendation system is usable or can be used to provide advice to students in selecting a major in SMK.

INTEGRATING GHRM AND INFORMATION SYSTEMS FOR SUSTAINABILITY IN OIL & GAS: 2019-2023 BIBLIOMETRIC STUDY

Adryan Rachman — 2024-04-22

In an era where ecological sustainability intersects with corporate accountability, this study pioneers the integration of Green Human Resource Management (GHRM) practices within the oil and gas sector, heralding a paradigm shift towards environmental stewardship and sustainable business operations. Through an incisive bibliometric analysis of data sourced from Scopus, ScienceDirect, and ResearchGate databases covering the years 2019-2023, this research meticulously examines a corpus of 52 journals, selectively narrowing down to 10 seminal works that underscore the critical nexus between organizational strategies and sustainable investment practices. Leveraging the analytical prowess of VosViewer for a nuanced exploration of the thematic landscape, this study illuminates the transformative role of GHRM in driving the oil and gas industry towards a future where environmental considerations are not merely an adjunct to business strategy but are embedded within the core operational ethos. The investigation transcends conventional academic discourse, positioning itself as a clarion call for the industry to reevaluate and realign its project management and investment frameworks in favor of sustainable development principles. The findings of this research are both a testament to and a roadmap for the integration of sustainability into the fabric of corporate strategy, highlighting innovation, strategic adaptability, and a steadfast commitment to green practices as indispensable to securing the sector’s long-term viability. This study not only contributes to the burgeoning field of GHRM but also sets a benchmark for future research, advocating for a symbiotic relationship between economic growth and environmental preservation. This groundbreaking work challenges the oil and gas sector to lead by example, embodying the principles of sustainability in every facet of its operations, thereby setting a new standard for corporate responsibility in the face of global ecological challenges.

THE RIGHT STEPS TOWARDS GRADUATION: NB-PSO SMART COMBINATION FOR STUDENT GRADUATION PREDICTION

Ahmad Hafidzul Kahfi — 2024-04-22

The current digital era demands a more innovative approach in predicting student campuses considering that campuses are not only important for students but also for lecturers, student guardians and higher education institutions. Previous studies have used various machine learning methods such as Decision Trees, Neural Networks, Support Vector Machines, etc. in these predictions. The problem that occurs is that even though various machine learning methods have been used, there are still limitations in the accuracy and efficiency of predicting student admissions, The problem in question can be given a real example of a case that occurred. So with this problem the aim is to develop a more effective methodology in predicting student permits, with recommendations from an intelligent combination of two computational techniques Naive Bayes (NB) and Particle Swarm Optimization (PSO). This research methodology includes data collection, NB model development and model partnership with PSO. Student graduation data is used in model testing with evaluation based on metrics such as accuracy and Area Under the Curve (AUC). The results showed a significant increase in accuracy to 86.94% from 83.30% and AUC value from 0.860 to 0.884 when using the combination of NB and PSO compared to NB without either. The integration of NB and PSO has been proven to increase effectiveness in classifying student graduation prediction cases. This research opens up opportunities for the practical application of technology in the education sector and emphasizes the importance of using effective optimization and feature selection techniques in improving prediction results.

SENTIMENT ANALYSIS OF CYBERBULLYING USING BIDIRECTIONAL LONG SHORT TERM MEMORY ALGORITHM ON TWITTER

Anisa Ika Safitri — 2024-04-22

Cyberbullying on social media such as Twitter is becoming an increasing social problem in today's society. Cyberbullying has a negative influence on mental health, increasing the risk of anxiety, sadness, and even suicide. The purpose of this research is to develop a model to classify tweets that contain or do not contain cyberbullying by applying the BiLSTM technique to sentiment analysis on Twitter. In this research, Word2Vec is used to weight each word in a tweet. The initial stage in this research is data collection with a total dataset of 47,692 tweets generated by Kaggle, preprocessing which consists of data cleaning, removing duplicates, case folding, tokenizing, stopword removal and lemmatization, classification and evaluation. This research uses the Bidirectional Long Short-Term Memory (Bi-LSTM) method and identifies patterns associated with bullying on social media. Testing uses Confusion Matrix and the results on classification show accuracy of 82.29%, precision of 82,04%, recall of 81,95% and F1-Score 81,89%. This sentiment analysis technique is expected to be the first step to combat and avoid cyberbullying on the Twitter platform. From several tests of existing reference algorithms, the classification accuracy performed includes having good performance.

DETECTION OF VEHICLE TYPE AND LICENSE PLATE WITH CONVOLUTIONAL NEURAL NETWORK MODEL YOLOV7

Suhartono Suhartono — 2024-04-22

This research was conducted in response to issues related to the efficiency and effectiveness of vehicle type and license plate detection. The increasingly congested traffic conditions and the expanding use of motor vehicles have posed challenges in traffic monitoring and regulation. Therefore, there is a need to develop a solution that can save time and resources while providing more comprehensive information in vehicle monitoring. This research implements the Convolutional Neural Network (CNN) algorithm with the latest YOLOv7 model from YOLO to detect vehicle types and vehicle number plates simultaneously to make it more efficient and effective, save time and resources, and provide more complete information. The research method used is Research and Development (R&D) with an experimental approach. The stages include image acquisition, labeling, dataset sharing, YOLOv7 model training, testing, prediction results, and conversion to text using Optical Character Recognition (OCR). The research results show that the ResNet34 model architecture achieves a total accuracy of 89.7% for 3x3 convolution layers and 88.6% for 5x5 convolution layers. The YOLOv5 architecture performs well on 3x3 convolution layers with an overall accuracy of 71.9%, and 58.3% for 5x5 convolution layers. However, the YOLOv7 and Mobilenet architectures tend to have lower accuracy, namely the Mobilenet architecture with a 3x3 convolution layer with a total accuracy of 63.4%, and 73.4% for the 5x5 convolution layer. Computing speed is also considered, with YOLOv5 and YOLOv7 having higher speeds than ResNet34 and Mobilenet. Tests were carried out in various lighting conditions, resulting in accurate detection of vehicle types and vehicle number plates of 90% in the morning, 85% in the afternoon and 77% at night. Overall, the system succeeded in recognizing objects with an accuracy of 84% from a total of 720 data tested, but the accuracy of converting vehicle number plates using OCR reached 22%. The results of this research demonstrate the performance and effectiveness of the YOLOv7 algorithm in detecting vehicle types and vehicle number plates, as well as providing insight into accuracy in various lighting conditions and OCR conversion.

CLASSIFICATION OF RICE PLANTS AFFECTED BY RATS USING THE SUPPORT VECTOR MACHINE (SVM) ALGORITHM

Nofie Prasetiyo — 2024-04-27

In the era of Indonesia's agrarian economy which is supported by the agricultural sector, rice plants play an important role in meeting food needs. However, pest attacks, especially field mice, can cause significant losses in rice production. To overcome this, this research proposes the use of the Support Vector Machine (SVM) algorithm with the Particle Swarm Optimization method in predicting rat pest attacks on rice plants. This research involves the process of collecting data from drone photos to identify affected agricultural land. The preprocessing stage involves changing colors from RGB to GRAY and zoom augmentation. Feature extraction is carried out using Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP). Testing was carried out involving the SVM/SVC model and performance evaluation was carried out using accuracy, precision and recall metrics. The preprocessing test results showed an increase in performance with training accuracy of 68.33%. However, the actual prediction on the original image results in a low accuracy of around 25%. However, image testing after involving the entire process, including preprocessing and model prediction, shows a higher level of accuracy, reaching around 90%.