JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer)

PREDICTION OF INHIBITOR BINDING AFFINITY AND MOLECULAR INTERACTIONS IN MPRO DENGUE USING MACHINE LEARNING

Venia Restreva Danestiara, Marwondo Marwondo, Nayla Nurul Azkiya — Mon, 03 Feb 2025 00:00:00 +0000

The dengue virus experiences rapid mutation and genetic variability, posing challenges in developing effective antiviral therapies. This study explores the prediction of binding affinities between potential antiviral drug inhibitors and the NS2B-NS3 protease of the dengue virus using machine learning models. Molecular docking simulations were conducted with AutoDock Vina to generate interaction data between viral proteins and ligands. The generated datasets were used to train several machine learning models, including Random Forest Regressor (RF Regressor), Support Vector Regression (SVR), and Extreme Gradient Boosting Regressor (XGBoost Regressor). The RF Regressor model demonstrated the highest accuracy in predicting binding affinities, measured through Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Pearson Correlation Coefficient (R). However, the XGBoost Regressor and SVR models showed better generalization in practical scenarios. This study highlights the potential of machine learning to optimize the drug discovery process and provides significant insights into antiviral drug development for dengue fever.

IMPLEMENTATION OF K-MEDOIDS METHOD FOR HEART DISEASE PREDICTION USING QUANTUM COMPUTING AND MANHATTAN DISTANCE

Mochamad Wahyudi, Dimas Trianda, Lise Pujiastuti, Solikhun Solikhun — Mon, 03 Feb 2025 00:00:00 +0000

Heart disease is a severe health condition characterized by dysfunctions in the heart and blood vessels, which can be fatal if not properly managed. Early detection and prediction of heart disease are crucial for understanding the prevalence and determining patients' quality of life. In this study, quantum computing is applied to enhance the performance of the K-Medoids method. A comparative analysis of these methods is conducted, focusing on their performance. The investigation utilizes a dataset of heart disease patient medical records. This dataset includes various attributes used to predict heart disease patterns. The dataset is tested using both the classical and K-Medoids methods with a quantum computing approach, employing Manhattan distance calculations. This study's findings reveal that applying quantum computing to the K-Medoids method results in clustering accuracy stability of 85%, equivalent to the classical method. Although there is no increase in accuracy, the quantum computing approach demonstrates potential improvements in data processing efficiency. These results highlight that the K-Medoids method with a quantum computing approach can contribute significantly to faster and more efficient medical data analysis. However, further research is needed for optimization and testing on more extensive and more diverse datasets.

OUTSOURCED EMPLOYEE RECRUITMENT DECISION SUPPORT SYSTEM WITH FUZZY TOPSIS INTEGRATED REST API METHOD

Asep Denih, Asep Saepulrohman, Febri Febriansyah — Mon, 03 Feb 2025 00:00:00 +0000

PT Dina Mika Muda Mandiri is a logistics and transportation company that is facing challenges in recruiting outsourced employees to meet the company's standards with complex assessment criteria. In overcoming this problem, the research developed a decision support system that is integrated with Rest API and the Fuzzy Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method. The system aims to improve the efficiency and accuracy of candidate selection by evaluating criteria such as interviews, knowledge, testing, curriculum vitae (CV), processing time, and salary. Two case studies were conducted involving 36 applicants for a website upgrade project and 24 applicants for an outsourced goods transit system. The results demonstrate that the decision support system integrated with Fuzzy TOPSIS significantly enhanced the selection process, improving accuracy by 91% for the website upgrade project and 97% for the goods transit system when compared to traditional human resource development (HRD) decision criteria. This demonstrates the system's effectiveness in aligning with HRD standards, making the recruitment process more effective, accurate and efficient. Future research should explore methods to refine the weighting of criteria and integrate expert opinions or more sophisticated machine learning algorithms to support more objective decision support systems in outsourcing employee recruitment.

DESIGNING USER EXPERIENCES IN CASUAL GAMES TO ENHANCE PRODUCT KNOWLEDGE

Eva Handriyantini, Stefanus Salem — Mon, 03 Feb 2025 00:00:00 +0000

The widespread adoption of mobile platforms has transformed the gaming industry, making casual games highly popular due to their accessibility via smartphones and tablets. Beyond entertainment, casual games now serve as effective educational and marketing tools for delivering product knowledge. This study explores how user experience (UX) design can enhance product education in casual games by focusing on game mechanics, UX principles, narrative engagement, and product placement. Using a Design-Based Research (DBR) approach, this study develops, tests, and refines interactive experiences to ensure the effective implementation of design elements. Testing with 50 participants showed a 30% improvement in product recall after playing, along with high satisfaction levels regarding game usability and engagement. Participants also demonstrated improved time management skills and emotional connection to the game content. The game integrates challenges and activities designed to build cognitive and emotional engagement. Artificial intelligence (AI) technology is utilized through Unreal Engine to create a realistic and immersive environment. By incorporating product information into engaging gameplay, the game serves as both an educational and entertainment tool. This research provides practical insights for game developers, marketers, and educators on integrating educational content into casual games. By leveraging AI, user testing, and advanced UX strategies, casual games can become effective tools for game-based marketing and education. This game significantly enhances product knowledge retention, user engagement, and practical skills.

SMART ATTENDANCE TRACKING SYSTEM EMPLOYING DEEP LEARNING FOR FACE ANTI-SPOOFING PROTECTION

Bani Nurhakim, Ahmad Rifai, Dian Ade Kurnia, Dadang Sudrajat, Ujang Supriatna — Mon, 03 Feb 2025 00:00:00 +0000

Conventional attendance systems face challenges in accuracy and efficiency, often vulnerable to spoofing and data manipulation. This study addresses these issues by developing a smart attendance system integrating Deep Learning-based facial recognition with anti-spoofing technology. The system ensures secure and reliable attendance authentication while automating and enhancing management processes. Utilizing a convolutional neural network (CNN) architecture, the system processes raw facial images directly without additional feature extraction, improving accuracy and efficiency. A novel training strategy, termed 50 Random Samples-30 Sub-epochs Count-1 Epoch, is introduced to optimize the training process. This strategy involves random sampling during each forward pass and grouping 30 passes as one epoch, enabling the use of complex CNN architectures and automatic dataset expansion. The system achieves 98.90% accuracy in identifying genuine attendance, maintaining a confidence level above 80%, significantly reducing spoofing risks and errors. This innovative solution has significant implications, particularly for educational institutions. It automates attendance tracking, minimizes manual effort, reduces errors, and supports disciplinary enforcement through accurate data. Moreover, its scalability allows for application across various environments, offering benefits to a wide range of institutions. By enhancing data accuracy and operational efficiency, this system sets a foundation for smarter, more reliable attendance management, strengthening administrative practices in education and beyond.

MULTIPLAYER ONLINE ROLE-PLAYING GAME VIRTUAL CLASSROOMS USING THE GAME DEVELOPMENT LIFE CYCLE METHOD

Mon, 03 Feb 2025 00:00:00 +0000

The COVID-19 pandemic has disrupted traditional education, forcing a shift toward online learning, which often lacks engagement and effectiveness. Existing virtual classroom methods struggle to sustain students' attention and motivation, leading to reduced learning outcomes. To address these issues, this study develops an innovative Virtual Classroom application based on Multiplayer Online Role-Playing Game (MORPG) technology. The goal is to provide a more interactive and immersive learning environment, enhancing engagement among students and lecturers. Using the Unity Game Engine, Photon Unity Networking (PUN), and Photon Voice libraries, this application transforms online classes into game-like experiences. The development followed the Game Development Life Cycle (GDLC) methodology, ensuring a structured and effective approach. Blackbox testing confirmed that all functions operated as intended, while usability testing with the System Usability Scale (SUS) among 30 users yielded an average score of 71.92, indicating a satisfactory experience. The results demonstrate the application's potential to make online learning more appealing and effective, contributing a novel solution for remote education challenges by integrating gaming elements into the learning process.

COMPARISON OF ENSEMBLE METHODS FOR DECISION TREE MODELS IN CLASSIFYING E. COLI BACTERIA

Alvin Rahman Al Musyaffa, Yoga Pristyanto, Nia Mauliza — Mon, 03 Feb 2025 00:00:00 +0000

Certain strains of Escherichia coli (E. coli) can cause serious illness, so identifying dangerous strains with high accuracy is a priority in supporting public health and food safety. However, traditional machine learning methods, such as Decision Trees, are often not robust enough to handle the complexity of biological data. This research presents a solution by systematically evaluating seven ensemble methods, namely Adaboost, Gradient Boosting, XGBoost, LightGBM, Random Forest, Bagging, and Stacking, using a dataset that includes 336 E. coli samples with eight biological features. These models are evaluated based on accuracy, precision, recall, and F1 score, with parameter optimization to obtain the best results. The results show that XGBoost is superior with accuracy, recall, and F1 score of 88% and precision of 87%, outperforming other methods. This research has the advantage of a comprehensive approach in comparing various ensemble methods simultaneously, accompanied by the application of confusion matrix-based evaluation to ensure the accuracy of the results. Additionally, the ensemble approach proved to be more effective in handling complex data patterns and reducing bias in bacterial strain classification. These findings provide a significant contribution, namely a practical framework for improving laboratory diagnostics and public health surveillance, with machine learning-based solutions that are faster, more reliable, and applicable for both industrial and clinical environments. This research expands understanding of the potential of ensemble methods in microbiological data classification and provides new directions for modern diagnostic technology.

CAUSAL MODELING OF FACTORS IN STUNTING USING THE PETER-CLARK AND GREEDY EQUIVALENCE SEARCH ALGORITHMS

Yohani Setiya Rafika Nur, Aminatus Sa’adah, Dasril Aldo, Bidayatul Masulah — Mon, 03 Feb 2025 00:00:00 +0000

Stunting is one of the nutritional problems that can hinder the growth and development process in toddlers. Untreated stunting can lead to fatal outcomes. Previous research on the factors that exist in the incidence of stunting mostly used multivariate analysis. Previous research on stunting factors has primarily used multivariate or correlation analyses. However, this study uniquely focuses on establishing causal relationships between these factors, a crucial step in improving early diagnosis for stunting prevention and treatment. The data used in this research was 83 data on stunting incidents and consisted of eight parameters. The purpose of this study is to model the causal relationship between factors that represent the incidence of stunting. This study uses two simple causal approaches, namely the Peter-Clark (PC) algorithm to obtain the initial concept of a graph model of the relationship between variables and the Greedy Equivalence Search (GES) algorithm to refine the model by obtaining the direction of the causal relationship. There are six bi-directed relationships that have been found, namely from food variables to support; maternal knowledge with sanitation; Height/Age and Weight/Age with Child Nutrition; height/age with weight/age and stunting. In addition, both algorithms in this study have successfully obtained a causal model, by comparing performance using directional and causal densities that the GES algorithm was able to identify a relationship of 0.66 compared to the PC algorithm.

DEVELOPMENT OF VT-UNUJA APPLICATION AS A WEBVR-BASED CAMPUS ENVIRONMENT INTRODUCTION MEDIA

Miftahul Huda, Fathorazi Nur Fajri, Maulidiansyah Maulidiansyah — Mon, 03 Feb 2025 00:00:00 +0000

Conventional campus introductions are often limited in providing an immersive experience to prospective students, especially for those who cannot attend in person. This encourages the need for technology-based solutions that can overcome these limitations. This research develops a WebVR-based VT-UNUJA application as a campus introduction media that offers an interactive experience with 360-degree panoramic image features, hotspot descriptions, navigation, and voice-over. The purpose of this research is to create an application that can increase user understanding of campus locations and facilities more efficiently and easily accessible. The test results show that this application is effective in improving user understanding, with a high level of satisfaction with the ease of use and interactivity of the application. The benefits of this research are to contribute in improving campus professionalism in presenting information digitally, as well as providing innovative alternatives for other educational institutions in supporting the orientation process for prospective students.

APPLYING K-MEANS CLUSTERING FOR GROUPING PAPUA’S DISTRICTS BASED ON POVERTY INDICATORS ANALYSIS

Yusriana Chusna Fadilah, Asrul Sani, Andrianingsih Andrianingsih — Mon, 03 Feb 2025 00:00:00 +0000

In the context of Indonesia's resource-rich development, poverty remains a major challenge, especially in Papua Province which has the highest poverty rate. Although Papua is rich in resources such as minerals, tropical forests, and biodiversity, challenges such as economic inequality, lack of infrastructure, and social conflict hinder economic and social progress. This research aims to implement the K-Means Clustering algorithm to cluster districts/cities in Papua based on poverty indicators, including the percentage of poor people, poverty line, average years of schooling, human development index, poverty depth index, poverty severity index, unemployment rate, and per capita expenditure. The research methodology includes data collection from the Central Statistical Agency (BPS), data processing through cleaning and transformation stages, and application of K-Means Clustering to determine the optimal cluster using the elbow method and silhouette score. The results show that the districts/cities in Papua can be grouped into two main clusters: C0, which indicates high poverty rates and C1, which indicates low poverty rates. This research is expected to provide a strategic foundation for the government to design more focused and effective development policies in reducing poverty in Papua.

OPTIMIZING MSME PRODUCT AUTHENTICITY VERIFICATION IN DECENTRALIZED MARKETS USING BLOCKCHAIN

Adnan Zulkarnain, Mukhlis Amien — Mon, 03 Feb 2025 00:00:00 +0000

Blockchain technology offers a solution for ensuring product authenticity in decentralized digital marketplaces. However, Micro, Small, and Medium Enterprises (MSMEs) face barriers such as limited infrastructure, high costs, and data interoperability challenges. This study develops a hybrid blockchain-based application architecture tailored to MSME needs, integrating on-chain and off-chain storage. Critical security data, such as product hashes, is stored on-chain, while non-sensitive data, like product descriptions, is managed off-chain using a cloud-based MySQL database. This design reduces storage costs and computational load while maintaining data integrity. Ethereum smart contracts manage product registration and verification, linked to QR code-based authentication for end-users. A realistic simulation environment using server-based infrastructure and cloud databases evaluated system performance, including transaction throughput, latency, resource utilization, and scalability. The results show significant improvements compared to conventional centralized methods, achieving a transaction throughput of 391 TPS for 1 million transactions while maintaining low latency and resource efficiency. This research addresses a theoretical gap by optimizing blockchain for small-scale decentralized markets, tackling resource limitations and interoperability issues unique to MSMEs. Practically, it provides a scalable and cost-effective solution for product authenticity verification, enhancing consumer trust and reducing counterfeiting in MSME digital markets. While real-world testing remains a limitation, the findings underline the system’s potential to support sustainable MSME digital marketplaces and build consumer confidence.

DENTAL CARIES SEVERITY DETECTION WITH A COMBINATION OF INTRAORAL IMAGES AND BITEWING RADIOGRAPHS

Jennifer Jennifer, Winni Setiawati, Gabriella Adeline Halim, Tony Tony — Mon, 03 Feb 2025 00:00:00 +0000

Dental caries is a multifactorial oral disease caused by plaque due to bacterial sugar fermentation. Quite a number of dentists have misdiagnosed caries due to the subjective nature of visual examination and radiograph in early-stage lesions. Thus, research on the implementation of deep learning technology is expected to improve the accuracy of diagnosis. However, caries detection with deep learning has accuracy problems. This problem makes researchers interested in developing a deep learning method that combines Faster R-CNN algorithm and texture feature extraction to more accurately detect carious teeth from bitewing radiography datasets and intraoral images. The overall performance of the model to detect the radiographic class was slightly better than the intraoral class. Overall, the classification accuracy of the model was 88.95% which is better than previous research that only used one or the other type of images. GLCM (Gray-Level Co-Occurrence Matrix) is effective in detecting contrast areas, but it still cannot specifically distinguish normal anatomical contrast from caries. The Faster R-CNN model learned well and was able to differentiate between each caries type and was successfully integrated with the GLCM matrix for radiographic image pre-processing to facilitate caries detection. This approach could have the potential of assisting dental professionals in reducing diagnostic errors and increasing patient care.

COMPARISON OF ACTIVATION AND OPTIMIZER PERFORMANCE IN LSTM MODEL FOR PURE BEEF PRICE PREDICTION

Dasril Aldo — Mon, 03 Feb 2025 00:00:00 +0000

One of the primary factors impacting the economy is the ability to forecast the prices of commodities such as beef. This paper aims to evaluate the effectiveness of various activation functions and optimization strategies when integrated into the LSTM (Long Short-Term Memory) architecture model in predicting the price of lean beef in Aceh. The data sample utilized was obtained from the Indonesian National Food Agency panel, which shows daily prices for beef within the time frame of July 14th, 2022, to July 31st, 2024. As for the conducted research, the process of preparation data preprocessing, partitioning data into training, validation and test sets and the actual execution of the LSTM model which was trained using four different types of activation functions: tanh, ReLU, sigmoid and PReLU together with three different optimizers: Adam, Nadam and RMSprop for 50, 70, 100 and 200 training iterations. The evaluation metrics employed were Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R-squared). The best performance was recorded at 200 epochs with the combination of PReLU activation function and Nadam optimizer, which had the best performance with RMSE = 2.56, MAPE = 0.65% and R² = 0.104. This combination was more effective than others since it depicted better overall performance in identifying complex non-linear relationships that existed in the price data. Further on, Nadam seems to have benefits in terms of allowing the model to converge faster and making the training more stable. This work stresses the selection of activation functions and optimization methods when building LSTM models aimed at forecasting prices of commodities with large volatility. It will be very helpful in developing better predictive models and decision-making processes in the agro-business. Another way to enhance predictive performance could be changing the model architecture or using different techniques, such as attention mechanisms.

ASPECT-BASED SENTIMENT ANALYSIS ON TWITTER TWEETS ABOUT THE MERDEKA CURRICULUM USING INDOBERT

Andi Wafda, Dhomas Hatta Fudholi, Jaka Nugraha — Mon, 03 Feb 2025 00:00:00 +0000

The curriculum has changed once again with the introduction of the Merdeka Curriculum to address learning loss in the education sector. Its implementation has elicited various responses, such as support for granting teachers the freedom to innovate, focusing on essential materials, offering diverse learning methods, and fostering student creativity. However, criticism has also arisen, including issues related to teachers’ lack of understanding, parents' concerns, and the increased workload on students due to numerous projects. To improve educational policies, an in-depth analysis of these responses is essential. This study aims to analyze public sentiment toward the Merdeka Curriculum by applying Aspect-Based Sentiment Analysis (ABSA) using data from Twitter. The research focuses on four main aspects: Teaching Modules (MA), Education Reports (RP), the Merdeka Teaching Platform (PMM), and the Strengthening of the Pancasila Student Profile Projects (P5). Data were collected using specific and relevant keywords for each aspect, followed by preprocessing, labeling, and filtering based on sentiment and aspect. The final dataset comprised 2,359 valid tweets. The ABSA model was developed using IndoBERT with fine-tuning, then tested and evaluated. The results showed that the aspect classification model achieved an accuracy of 97%, F1 score of 97%, recall of 97%, and precision of 97%. Meanwhile, the sentiment classification model achieved an accuracy of 85%, F1 score of 85%, recall of 85%, and precision of 85%. This ABSA model is expected to assist in monitoring public responses and provide valuable insights for policy development, particularly within the context of the Merdeka Curriculum.

COMBINATION OF LEARNING VECTOR QUANTIZATION AND LINEAR DISCRIMINANT ANALYSIS FOR TEA LEAF DISEASE CLASSIFICATION

Mutasar Mutasar, Chaeroen Niesa — Wed, 12 Feb 2025 00:00:00 +0000

Tea farming, one of the key pillars of Indonesia's economy, faces productivity challenges due to diseases affecting tea leaves. Manual identification of tea leaf diseases requires significant time and cost, making an automated solution necessary. This research develops an innovative model for classifying tea leaf diseases by synergizing Learning Vector Quantization (LVQ) and Linear Discriminant Analysis (LDA). By leveraging LVQ’s prototype-based classification and LDA’s dimensionality reduction, the model ensures accurate and efficient disease identification. During preprocessing, tea leaf images were converted to the CIELAB color space to enhance segmentation using Otsu’s Thresholding. Features such as Mean Color and texture attributes based on Gray Level Co-occurrence Matrix (GLCM) were extracted, reduced via LDA, and classified using LVQ. Tested on five tea leaf disease classes, the model achieved 94.1% accuracy. This performance underscores its potential to significantly assist farmers in early detection and management of tea leaf diseases, while also providing researchers with a robust tool for advancing agricultural technology.

BEYOND ALGORITHMS: AN INTEGRATED APPROACH TO FAKE NEWS DETECTION USING MACHINE LEARNING TECHNIQUES

Bimantyoso Hamdikatama — Wed, 12 Feb 2025 00:00:00 +0000

The internet has become a major source of information, but it also facilitates the rapid spread of fake news, which can significantly influence public opinion and social decisions. While various techniques have been developed for detecting fake news, many studies focus on individual algorithms, which often result in suboptimal performance. This study addresses this gap by comparing machine learning models, including Support Vector Classification (SVC), XGBoost, and a Stacking Ensemble that combines both SVC and XGBoost, to determine the most effective approach for fake news detection. Text preprocessing was performed using IndoBERT, which provides context-aware and semantically rich text representations specifically for the Indonesian language. The evaluation results demonstrate that the Stacking Ensemble outperforms the individual models, achieving an accuracy of 82%, compared to 79% for XGBoost and 78% for SVC. This superior performance is attributed to the complementary strengths of the base models: SVC excels in handling high-dimensional data, while XGBoost effectively manages imbalanced datasets and captures complex feature interactions. The use of IndoBERT further enhances model performance by improving text representation through contextual embeddings. These findings highlight the effectiveness of ensemble learning in enhancing predictive performance and robustness for fake news detection, demonstrating the potential of combining different machine learning techniques with advanced preprocessing methods to achieve more reliable results.

INTELLIGENT SYSTEM TO DETERMINE THE BEST LECTURER USING ADDITIVE RATIO ASSESSMENT ALGORITHM

Wahyudi Wahyudi, Budy Satria, Lutfil Khairi — Wed, 12 Feb 2025 00:00:00 +0000

The quality of a lecturer's performance is one of the keys to institutional success that must be continuously improved. The performance assessment of lecturers in the Informatics study program of the Faculty of Information Technology, Andalas University faces obstacles in processing quantitative and qualitative data so that it is vulnerable to subjectivity including research productivity, teaching effectiveness, contributions to community service and additional activities. In addition, limitations in a systematic evaluation system result in unfairness and lack of transparency in the decision-making process. The research objective is to create a technology-based approach by applying the Additive Ratio Assessment method based on a Decision Support System. The ARAS method was chosen because it is able to determine effective final results based on multiple criteria that have been determined. The application of the ARAS method consists of 5 stages, namely determining the decision matrix, normalizing the decision matrix, weighting the normalization results, determining the optimum function value and ranking results. The results obtained are alternative data consisting of A1,A2,A3,A4,A5,A6,A7,A8,A9,A10,A11 and 8 criteria and weighting, namely the last education (10%), functional position (15%), certification (20%), number of publications (15%), author order (15%), publication index quality (10%), research grants (10%) and PkM (5%). The ranking results with the highest value in order 1-5 are 0.113875, 0.109785, 0.104235, 0.099005, 0.094715. The final conclusion of this research is that the ARAS method is able to prove the best lecturer assessment to be more efficient, transparent and subjective to be applied in the Andalas University Informatics study program.

DESIGN OF CONTROL SYSTEM AND TEMPERATURE IN COFFEE DRYER ARDUINO BASED AUTOMATIC USING FUZZY

Tue, 18 Feb 2025 00:00:00 +0000

The coffee bean drying process is a crucial stage in ensuring the final quality of coffee products. Conventional drying methods, which rely on sunlight, face several challenges, such as dependence on weather conditions and prolonged drying times. This study proposes the design of a control and temperature system for an automatic coffee dryer based on the Arduino Mega 2560, aimed at enhancing the efficiency and consistency of the drying process. The system utilizes a semi-enclosed drying technology equipped with DHT22 temperature and humidity sensors, controlled by Arduino-Uno and Fuzzy Logic. This control system monitors temperature and humidity in real-time, maintaining the drying conditions at 55°C and 15% RH. If the temperature or humidity exceeds the set limits, the system activates an LED and buzzer alarm, indicating that the drying process has reached optimal conditions. The prototype was tested under various conditions, and the results demonstrate that the system has a high accuracy level in controlling temperature and humidity, significantly accelerating the drying process compared to traditional methods. By implementing this technology, the coffee industry in Indonesia is expected to achieve the Coffee Drying Operational Standards in accordance with SNI, maintain flavor quality, optimize the use of drying land, and reduce drying duration. This development offers an innovative solution that can enhance the quality and productivity of coffee processing, providing significant economic benefits to farmers and coffee industry stakeholders.

UEQ-BASED EVALUATION OF USER EXPERIENCE: A CASE STUDY ON ENGLISH READING WEBSITES DEVELOPMENT

Tue, 18 Feb 2025 00:00:00 +0000

Technology has become essential in education, requiring specialized instructional media tailored to each subject. This study develops a web-based learning system designed specifically for English education, integrating both passive and interactive learning approaches to enhance engagement and effectiveness. This study contributes to showing that passive-interactive English learning websites can help users in both the advanced and non-advanced categories. However, before the system is publicly accessible, system testing focusing on user experience is important to achieve the intended learning objectives. The current study employed the User Experience Questionnaire (UEQ) framework, consisting of 26 questions across the categories of Attractiveness, Pragmatic, and Hedonic. The UEQ framework has been well-validated and widely used in various studies and professional industries for assessing user experience, ensuring that the user experience scores are objective. The findings showed that the system scored lowest in the “Hedonic” category due to a low “Novelty” score of 1.344, which was 31% lower than the highest score “Attractiveness” at 1.962. Interface that visually appealing and offers fresh, dynamic content, users are more likely to stay engaged. However, the overall average score across the three UEQ categories for the English learning system was 1.823, indicating a good user experience.

DESIGN OF FIRE EXTINGUISHER ROBOT USING IOT WITH ANDROID APPLICATION CONTROL

Tue, 18 Feb 2025 00:00:00 +0000

Fire is an unsupervised incidental disaster. This disaster has a detrimental impact on living and non-living things in the surrounding environment. This study was conducted to design an intelligent firefighting robot using Arduino Mega 2560 and Android-based IoT technology. This firefighting robot uses several Node MCU ESP8266 components as additional devices to connect to wifi. The L298N module regulates the speed and direction of the DC motor rotation, followed by the L9110 fan as hardware to extinguish the fire. The mobile robot prototype uses a DC motor as its driver. In addition, an Android application has been programmed to control the firefighting robot. This application has features that allow the robot to move in various directions and adjust the fan speed when extinguishing fires, all through an internet network connection. The study results showed that the application can be connected within a distance of 1-8 meters with good network quality. The test results showed that at a distance of 1-28 cm, the fan worked very well according to its function, and the Android application also worked optimally. In that range, the fan can extinguish the simulated fire source. The results of this study obtained a new approach to autonomous fire detection and extinguishing using IoT and robotic technology. In addition, it is able to integrate an Android-based IoT controller to enable remote control with real-time monitoring to overcome problems in previous research.

THE IMPACT OF WORD EMBEDDING ON CYBERBULLYING DETECTION USING HYBIRD DEEP LEARNING CNN-BILSTM

Moh. Hilman Fariz, Erwin Budi Setiawan — Tue, 18 Feb 2025 00:00:00 +0000

Cyberbullying can be perpetrated by anyone, whether children or adults, with the primary aim of belittling or attacking specific individuals. Social media platforms like X (formerly Twitter) often serve as the primary medium for cyberbullying, where interactions frequently escalate into retaliatory attacks, intimidation, and insults. In detecting these actions, short tweets are often difficult to understand without context, making specialized approaches like word embedding important. This research uses GloVe feature expansion, utilizing a corpus generated from the IndoNews dataset containing 127,580 entries to enhance vocabulary understanding in tweets that include the use of Indonesian language in both formal and informal forms. This data was then classified using the Hybrid Deep Learning method, which combines Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) with used 30,084 tweets taken from platform X as the dataset. The analysis results show that the application of expansion features using GloVe can improve the performance of the BiLSTM-CNN hybrid model, with the highest accuracy reaching 83.88%, an increase of +3.65% compared to the hybrid model without GloVe. This research successfully detected cyberbullying on platform X, making a significant contribution to efforts to create a safer and more positive social media environment for users.

OPTIMIZING THE KNN ALGORITHM FOR CLASSIFYING CHRONIC KIDNEY DISEASE USING GRIDSEARCHCV

Muhammad Rahmansyah Siregar, Dedy Hartama, Solikhun Solikhun — Thu, 20 Feb 2025 00:00:00 +0000

Chronic Kidney Disease (CKD) is a progressive condition that impairs kidney function and cannot be cured. Early detection is crucial for effective management and therapy. However, diagnosing CKD is challenging as patients often have comorbidities such as diabetes, hypertension, or heart disease, which complicate diagnosis and treatment. Accurate classification methods are essential for early detection. K-Nearest Neighbor (KNN) is a classification algorithm that groups data based on feature similarity. K-NN is an algorithm that is resistant to outliers, easy to implement, and highly adaptable. It only requires distance calculations between data points and does not involve complex parameters. However, its performance depends on hyperparameters such as the number of neighbors (k), weighting, and distance metric. Incorrect hyperparameter selection can lead to overfitting, underfitting, or reduced accuracy. To address these issues, GridSearchCV is used to optimize KNN by systematically selecting the best hyperparameters, ensuring improved accuracy and reduced overfitting. This optimization enhances the model’s reliability in early CKD detection compared to other methods. This study aims to determine the optimal KNN parameters for CKD classification using GridSearchCV. The results show 8.05% accuracy improvement and reduction in overfitting, with the prediction gap between training and testing decreasing from 6% to only 1.15%. These enhancements contribute to more reliable CKD diagnosis, enabling accurate early detection and better clinical decision-making.

DEVELOPMENT OF GRAPH GENERATION TOOLS FOR PYTHON FUNCTION CODE ANALYSIS

Fri, 21 Feb 2025 00:00:00 +0000

The increasing complexity of programs in software development requires understanding and analysis of code structure, especially in Python, which dominates machine learning and data science applications. Manual static analysis is often time-consuming and prone to errors. Meanwhile, static analysis tools for Python, like PyCG and Code2graph, are still limited to generating call graphs without including dependency and control flow analysis. This research addresses these shortcomings by proposing the development of a web-based tool that integrates the generation of function call graphs, function dependency graphs, and control flow graphs using Abstract Syntax Tree (AST), Graphviz, and Streamlit. With an iterative SDLC methodology, this tool was developed gradually to visualize Python function code as a heterogeneous graph. Evaluation of 11 Python function codes showed a success rate of 95.45% in analyzing and visualizing Python function codes with various levels of complexity. The limitations of Graphviz present an opportunity for future research to focus on improving scalability and Python code analysis.