Jurnal Ilmu Komputer dan Informatika

Implementation of the Waterfall Method in Designing a Health Center Information System

Andriansyah Andriansyah — 2025-07-20

Information systems develop along with the rapid development of information technology and have proven to play a role in various activities, such as storing, managing and distributing information by creating a website in its application. Puskesmas is a health service with the aim of providing basic health services to the community, and providing socialization about health. The problem that occurs is how the results of the application of the data collection method on the design of the puskemas information system and how the application of the implementation of the waterfall method designs a website-based puskesmas information system. the results of data management obtained by the puskesmas information system there are 4 processes consisting of profile data management, doctor data management, service data management, user data management. The application of the waterfall method in designing a web-based puskesmas information system can solve problems that occur and provide solutions in disseminating information quickly and accurately to the community. The design of the puskesmas information system has five (5) main designs, namely: 1). The system can display the visitor page when the visitor activates the application on the browser, 2). The system can receive and process data when the user performs CRUD (create, Read, Update, Delete) actions, 3). The system can verify login data and display an error message if the user enters the wrong login data, 4). The system can add

Evaluation of Transfer Learning-Based Convolutional Neural Networks (InceptionV3 and MobileNetV2) for Facial Skin-Type Classification

Naufal Hafizh Muttaqin — 2025-08-05

Manual classification of facial skin types often suffers from subjectivity and inconsistency due to reliance on human expertise. Accurate identification of skin types is crucial for selecting appropriate skincare solutions. This study evaluates the performance of two transfer-learning-based Convolutional Neural Networks (CNNs), InceptionV3 and MobileNetV2, for classifying facial skin types into four categories: normal, oily, dry, and acne-prone. A total of 1,733 facial images were collected from Kaggle and Roboflow and split into training, validation, and testing sets with a 70:20:10 ratio. Preprocessing involved normalization, augmentation, and resizing based on each model’s input size. Both models were fine-tuned and evaluated using accuracy, precision, recall, and F1-score metrics. InceptionV3 achieved the highest accuracy of 90.12% and a macro F1-score of 89.47%, particularly excelling in identifying normal and acne-prone skin. MobileNetV2 reached 81.15% accuracy and performed well on dry skin types. Confusion matrices and evaluation on new, unseen data confirmed the models’ generalization capabilities, though misclassifications still occurred among visually similar classes. These findings suggest that CNNs with transfer learning provide a robust foundation for developing AI-assisted facial skin-type classification systems, offering potential integration into dermatological applications.

Fire Detection Using Logistic Regression with GLCM, RGB Ratio, RGB Intersection, and Color Moments

Pieter Dickens — 2025-08-30

Fires pose a significant threat to human safety and property, particularly in densely populated urban environments where rapid and accurate early detection is critical. This study proposes an automated fire detection system based on computer vision and Logistic Regression classification, utilizing a combination of texture and color-based features to improve detection performance. The proposed approach integrates Gray-Level Co-occurrence Matrix (GLCM), RGB Ratio, RGB Intersection, and Color Moments to extract discriminative features from fire and non-fire images. The dataset, obtained from Kaggle, was preprocessed through HSV-based color segmentation to isolate candidate fire regions before manual annotation. The extracted features were then used to train a Logistic Regression model with hyperparameter tuning of the max_iter parameter to achieve optimal convergence. Experimental results show that the proposed model achieved an accuracy of 86% and a recall of 84% on the training dataset, and an accuracy of 87% with a recall of 82% on the test dataset. Despite these promising results, some false negatives were observed, indicating the need for further refinement to improve sensitivity. Comparative evaluation with a Convolutional Neural Network (CNN) demonstrated that the Logistic Regression approach achieved higher average processing speed, reaching up to 16.2 FPS for video input, compared to 11 FPS for CNN, making it more suitable for real-time applications. Overall, the integration of multi-feature extraction with Logistic Regression offers a balance between accuracy and computational efficiency for early fire detection in real-world scenarios.

Enhancing Fashion Product Sales Segmentation Using Random Forest with SMOTE and Hyperparameter Optimization

Guntur Tri Atmaja — 2025-08-30

The rapid expansion of the fashion e-commerce sector has intensified the need for accurate sales segmentation to support targeted marketing and efficient inventory management. This study proposes a robust methodology for classifying fashion product sales into three categories: high-selling, moderately-selling, and low-selling, using the Random Forest algorithm integrated with the Synthetic Minority Over-sampling Technique (SMOTE) and hyperparameter optimization. A real-world dataset comprising over 20,000 product records from an online marketplace was preprocessed through missing-value handling, categorical encoding, and numerical feature standardization. Class labels were generated using quantile-based segmentation of sales volume, followed by class balancing with SMOTE. The Random Forest model was tuned using RandomizedSearchCV and evaluated through accuracy, precision, recall, F1-score, and Receiver Operating Characteristic–Area Under Curve (ROC-AUC) metrics. Experimental results demonstrate strong predictive performance, achieving an accuracy of 90.43%, macro-precision of 90.60%, macro-recall of 90.45%, macro-F1 of 90.50%, and macro ROC-AUC of 0.9783. Feature importance analysis revealed that price, category, and customer ratings were the most influential predictors of sales segmentation. These findings validate the effectiveness of ensemble learning combined with class imbalance handling for multi-class classification in retail datasets. From a scientific perspective, this research contributes to the literature by presenting a reproducible, data-driven framework for product segmentation in heterogeneous and imbalanced datasets. Practically, the proposed approach can guide fashion retailers in refining pricing strategies, optimizing marketing campaigns, and improving inventory decisions in competitive online marketplaces. The methodology is adaptable to other e-commerce domains, offering broader implications for business intelligence and predictive analytics.

Teachers’ and Students’ Perspectives on the Ethical Use of Artificial Intelligence in Vocational Education using Technology Acceptance Model Approach

Nathan Christianto Kekado — 2025-08-30

Artificial Intelligence (AI) is increasingly integrated into educational settings, offering benefits such as improved efficiency, personalization, and student engagement. However, its adoption also raises ethical concerns that require careful consideration. This study investigates teachers’ and students’ perspectives on the ethical use of AI in vocational education, employing the Technology Acceptance Model (TAM) extended with Ethical Awareness, Trust, and Subjective Norms. A quantitative research design was applied, supported by interviews for triangulation. Data were collected from 60 students and 5 teachers in the Computer and Network Engineering program at SMK Negeri 2 Salatiga, Indonesia, all of whom had prior AI usage experience. The results indicate that Ethical Awareness significantly influences Attitude Toward Use (p = 0.002, t = 3.070), Behavioral Intention (p < 0.001, t = 6.175), and Perceived Usefulness (p < 0.001, t = 4.330). Perceived Ease of Use was found to have a positive effect on Behavioral Intention (p = 0.004, t = 2.913). Trust exhibited a strong relationship with both Actual Use (p < 0.001, t = 3.543) and Attitude Toward Use (p = 0.009, t = 2.621). Reliability testing showed Cronbach’s Alpha values above 0.70 for all key constructs, with Average Variance Extracted (AVE) values exceeding 0.50, indicating strong internal consistency and validity. These findings emphasize that ethical awareness and trust are critical determinants in fostering AI adoption in education. The study provides actionable insights for policymakers, educators, and technology developers to design training programs and guidelines that address ethical considerations, thereby ensuring responsible and sustainable AI integration in educational environments.

Comparative Analysis of Gaussian Naïve Bayes and Categorical Naïve Bayes Algorithms with Laplace Smoothing in COVID-19 Detection

Dila Saputra — 2025-08-30

In January 2020, it was confirmed that COVID-19 can be transmitted from human to human through the upper respiratory tract with a high infection rate. The number of COVID-19 cases worldwide continued to increase rapidly through close contact, droplets, and airborne transmission. In response, governments and the WHO implemented preventive measures, including COVID-19 treatment preparation, increased emergency healthcare capacity, and patient screening. Early detection of COVID-19 became crucial in taking action, providing treatment, and protecting others. In the Naïve Bayes algorithm, a potential issue arises with the possibility of zero probabilities for some features or attributes in the COVID-19 prediction training data. Therefore, Laplace Smoothing is used to address this problem. This study aims to compare the average accuracy rates of Gaussian Naïve Bayes and Categorical Naïve Bayes algorithms using different proportions of training data but the same testing data for COVID-19 detection. The methods used in this research are Gaussian Naïve Bayes and Categorical Naïve Bayes with Laplace Smoothing implemented using the Python library called scikit-learn. The research results show that the Gaussian Naïve Bayes algorithm without Laplace Smoothing has an average accuracy of 0.902165, while with Laplace Smoothing, it has an average accuracy of 0.973448. For the Categorical Naïve Bayes algorithm, without Laplace Smoothing, it has an average accuracy of 0.983864, while with Laplace Smoothing, it has an average accuracy of 0.984273. In conclusion, Laplace Smoothing plays a significant role in improving the average accuracy of Naïve Bayes algorithms. Categorical Naïve Bayes achieves the highest average accuracy of 0.9840685 (with and without Laplace Smoothing), while Gaussian Naïve Bayes achieves 0.947549 (with and without Laplace Smoothing). Categorical Naïve Bayes has a higher average accuracy compared to Gaussian Naïve Bayes.

Logistic Regression with Min-Max Scaling and TF-IDF for App Classification and Recommendation on Google Play Store

Calista Anindita — 2025-08-30

In the rapidly evolving mobile application ecosystem, enhancing user experience on the Google Play Store has become a critical challenge due to the vast number of available applications. This study proposes an integrated approach combining Logistic Regression, Min-Max Scaling, and the Term Frequency–Inverse Document Frequency (TF-IDF) Vectorizer to classify applications and generate personalized recommendations. The dataset, obtained from the Google Play Store, includes numerical features such as ratings, size, and installs, as well as textual data from user reviews. Min-Max Scaling was applied to normalize numerical attributes, ensuring balanced feature contributions during model training. TF-IDF was employed to convert textual reviews into meaningful numerical representations, enabling the model to capture the semantic importance of terms. The classification and recommendation system was evaluated using accuracy, precision, and recall as performance metrics. Experimental results demonstrated a substantial improvement compared to the baseline model, with accuracy, precision, and recall reaching 99.8%, compared to the previous 22.8% baseline performance. The system effectively recommended relevant applications based on user preferences, as measured through cosine similarity in feature space. These results indicate that the proposed method not only improves classification accuracy but also enhances the quality of app recommendations, thereby significantly improving user experience. The findings contribute to the field of computer science by demonstrating an effective integration of feature scaling and text vectorization into a classical machine learning model, offering a scalable and interpretable solution for large-scale recommendation systems in digital marketplaces. This approach can be further adapted to other domains requiring hybrid processing of numerical and textual data for predictive analytics.