Jurnal Nasional Teknik Elektro dan Teknologi Informasi

Sentiment Analysis of IKD Application Reviews on Play Store Using Random Forest

2025-09-26T08:34:09+07:00

The rapid growth of digital applications in population administration services has increased the importance of sentiment analysis to understand user perceptions more deeply. This study focuses on the Digital population identity (Identitas Kependudukan Digital, IKD), a digital identity application developed by the Indonesian government. It aims to classify user reviews of the IKD application into positive, neutral, and negative sentiments using the random forest algorithm. The dataset consisted of 28,134 user reviews from the Google Play Store, including usernames, review texts, timestamps, and star ratings. The research stages included data preprocessing, labeling, handling missing values, and text processing (cleansing, tokenizing, stopword removal, and stemming). The data were divided into 80% training and 20% testing sets. The best-performing model used the parameters: max_depth=None, max_features=log2, min_samples_leaf=1, min_samples_split=2, and n_estimators=300, achieving an average accuracy of 83.78%. To address class imbalance, the synthetic minority oversampling technique (SMOTE) was applied, resulting in improved performance with an accuracy of 86.29%. Evaluation metrics before SMOTE showed 83.85% accuracy, 80.40% precision, 83.85% recall, and 81.73% F1 score. After SMOTE, precision increased to 81.22%, while accuracy and recall slightly decreased to 80.86%, with an F1 score of 81.03%. Furthermore, sentiment trend analysis using N-gram techniques (unigram, bigram, trigram) was conducted to identify frequently mentioned topics and user concerns. These insights support the research objective of guiding application improvements aligned with user needs and enhancing the overall digital service experience.

Classification of Rice Diseases Using Leaf Image-Based Convolutional Neural Network (CNN)

2025-09-26T08:34:07+07:00

Rice diseases significantly impact agricultural productivity, making classification models essential for accurately distinguishing rice leaf diseases. Various classification models have been proposed for image-based rice disease classification; however, further performance improvement is still required. This study proposes the use of a convolutional neural network (CNN) to classify rice diseases based on leaf images. The dataset used in this study included leaf images categorized into four conditions: leaf blight, blast, tungro, and healthy. In the initial stage, data preprocessing was conducted, including resizing, augmentation, and normalization. Following preprocessing, a custom CNN architecture was developed, consisting of four convolutional layers, four pooling layers, and three fully connected layers. Each convolutional layer employed a 3 × 3 kernel with a stride of 1 and ReLU activation, while the pooling layers used max pooling with a 3 × 3 kernel and a stride of 2. Using a batch size of 32 and the Adam optimizer, the best test performance was achieved with 100 training epochs and a learning rate of 0.0002, resulting in a training accuracy of 0.9930, a loss of 0.0221, and a test accuracy of 0.9647. Model evaluation demonstrated a balanced performance across precision, recall, and F1 score, each achieving 0.9647, indicating highly effective classification without bias toward any specific class. These findings suggest that the simplified CNN model can deliver competitive classification performance without the need for complex architectures or additional enhancement techniques. The proposed CNN model outperformed existing CNN architectures, such as Inception-ResNet-V2, VGG-16, VGG-19, and Xception.

Interpretable Machine Learning for Job Placement Prediction: A SHAP-Based Feature Analysis

2025-09-26T08:34:04+07:00

Predictive modeling is important in analyzing graduates’ job outcomes, especially in forecasting job placements based on academic performance and courses. This study aims to improve predictive accuracy and interpretability in job placement classification using advanced machine learning models and SHapley Additive exPlanations (SHAP) analysis. Utilizing a dataset containing graduates’ academic records, including course grades, grade point average (GPA), and internship duration, this research employed several classification models, including decision tree, random forest, extreme gradient boosting (XGBoost), light gradient-boosting machine (LightGBM), CatBoost, and logistic regression. Evaluation metrics showed that most models achieve 92% precision, 92% recall, and 92% F1 score, with an accuracy of 85%, while logistic regression excelled with 100% recall, 96% F1 score, and 92% accuracy. SHAP analysis identified key features such as Administration, Computer Organization, Information Systems, Entrepreneurship, Professional Ethics, and Web Programming as the most influential in predicting job placement. Other significant contributors include Introduction to Information Technology, Software Engineering II, and Data Mining, although with relatively lower influence. Extracurricular activities and internship experiences were also found to be influential factors, highlighting the importance of academic and nonacademic elements in shaping graduates’ career prospects. These findings highlight and emphasize the need to provide students with certain academic courses to better prepare them for the job market. These findings emphasize the importance of interpretable machine learning models in career forecasting, enabling educational institutions to optimize curriculum design and enhance graduates’ employability. Future research should explore feature selection techniques, temporal analysis, and personalized recommendation systems to refine predictive accuracy.

Comparison of Sine-Cosine and Bat Algorithm for Distributed Generation Placement

2025-09-26T08:34:01+07:00

The enhancement of electricity distribution is a crucial factor in supporting sustainable development and reducing energy access inequality. To ensure the reliability and stability of energy systems, the integration of distributed generation (DG) has a significant role. Numerous studies have explored optimal DG placement using metaheuristic methods. The study evaluated the performance of both algorithms based on key indicators, including voltage profile improvement and power loss reduction, under normal load conditions and under a 10% load increase to simulate future demand growth. The methods employed were the sine-cosine algorithm (SCA) and the bat algorithm (BA). By comparing these two methods, this study aims to optimize the placement and sizing of DG units, with a case study based on the IEEE 9 bus system configuration. Load flow analysis was performed using Electric Transient Analysis Program (ETAP) software to validate the effectiveness of optimized DG placement under various scenarios. Key performance indicators, namely losses reduction and improvement of voltage profile, were evaluated to determine the relative strengths of each algorithm. The results show that both SCA and BA are effective in optimizing DG implementation. Specifically, SCA achieved reductions in active power losses by up to 85% and reactive power losses by 93%, outperforming BA in certain scenarios. Both algorithms enhance system reliability and stability. These findings highlight the potential of metaheuristic algorithms to address the challenges of modern energy systems and contribute to the broader goal of developing sustainable power systems.

A Machine Vision-Based Anthropometric System for Measuring Human Head Circumference

2025-09-26T08:33:57+07:00

This study aims to develop an automated anthropometric system based on machine vision, integrated into a medical cyber-physical system (MCPS), to measure human head circumference. Head circumference is a critical parameter in growth monitoring, particularly for detecting abnormalities such as microcephaly and macrocephaly, which can affect cognitive development and overall health. To address this challenge, the study proposed an anthropometric system that enabled automated, accurate, and contactless measurements, accessible in real-time by healthcare professionals. The system was designed using a machine vision approach, incorporating object detection technology and elliptical model-based perimeter estimation to determine head circumference noninvasively. A 1,920 × 1,080-pixel (1080p) camera operating at 30 fps with a 60° field of view was mounted on a three-axis motion mechanism driven by stepper motors to automatically capture frontal and side views of the head. The measurement process began with head detection and bounding box adjustment to obtain head width parameters. Euclidean distance was used for measurement, followed by elliptical geometry modeling to estimate head circumference. Experimental results showed the lowest error rate of 2.29% at a distance of 50 cm under 300 lux lighting conditions. Performance evaluation using a confusion matrix yielded an accuracy of 92.8%, precision of 100%, recall of 97.5%, and F score of 98.7%. The proposed system provides an effective solution for healthcare professionals to perform growth screening quickly, accurately, and safely. It also supports remote healthcare services, particularly in areas with limited access to medical facilities.

Attack Detection in IoT Networks Using Hybrid Feature Selection and Bayesian Optimization

2025-09-26T08:33:54+07:00

Machine learning (ML)-based attack detection is a promising alternative for addressing cybersecurity threats in Internet of things (IoT) networks. This approach can handle various emerging attack types. However, the growing volume of data and the reliance on default parameter values in ML algorithms have led to performance degradation. This study proposed a hybrid feature selection method combined with Bayesian optimization to improve the effectiveness and efficiency of attack detection models. The hybrid feature selection method integrated correlation-based filtering, which aimed to rapidly remove highly correlated features, and feature importance, which aimed to select the most influential features for the model. In addition, Bayesian optimization was employed to efficiently identify the optimal parameter values for lightweight and robust ML algorithms suitable for IoT networks, namely decision tree and random forest. The constructed model was then evaluated using the latest attack dataset, CICIoT2023, which consists of seven types of attacks: DDoS, DoS, Mirai, spoofing, reconnaissance, web-based attacks, and brute force. The evaluation results showed that the hybrid feature selection technique produced a more efficient model compared to several single feature selection methods by selecting 5 out of 46 features. Furthermore, Bayesian optimization successfully identified the optimal parameter values, improving model performance in terms of accuracy, precision, recall, and F1 score up to 99.74%, while reducing computational time by as much as 97.41%. Based on these findings, the proposed attack detection model using hybrid feature selection and Bayesian optimization can serve as a reference for implementing cybersecurity solutions in IoT networks.

Sustainable Generation and Transmission Expansion Planning Using MOPSO-BPSO in Electrical Grid

2025-09-26T08:33:52+07:00

As of 2023, approximately 85% of power plants operating in South Sulawesi relied on fossil fuels, such as coal, gas, and oil. To meet the increasing demand for electricity while reducing carbon emissions, it is essential to integrate renewable energy sources into the power system. Renewable energy not only helps conserve fossil fuels but also supports global environmental sustainability. South Sulawesi possesses significant hydro potential, offering opportunities to develop both small and large-scale hydroelectric power plants (pembangkit listrik tenaga air, PLTA). This study employed a multi-objective particle swarm optimization (MOPSO) approach to develop optimal scenarios for generation expansion planning (GEP), and binary particle swarm optimization (BPSO) to determine the necessary transmission expansion planning (TEP). The planning process was supported by long-term load forecasting using the moving average method based on historical electricity demand data in South Sulawesi. Results showed that the proposed integrated GEP and TEP optimization framework successfully identified an optimal scenario maximizing renewable energy used while ensuring transmission reliability. By 2030, PLTA is projected to contribute 67.9% of total electricity generation. Meanwhile, steam-fired power plants (pembangkit listrik tenaga uap, PLTU) become the mainstay with capacities reaching 437.5 MW. To support this scenario, nine new transmission lines are needed, along with the expansion of 25 existing lines to accommodate increased power flow within the interconnection system.