Public Research Journal of Engineering, Data Technology and Computer Science

Performance Comparison of Supervised Learning Algorithms in Heart Disease Risk Classification

2025-06-09T08:06:21+00:00

Heart disease is one of the leading causes of death worldwide, so early diagnosis is essential for effective treatment and prevention. This study evaluates the performance of five machine learning algorithms, namely Decision Tree, Naive Bayes, K-Nearest Neighbors (K-NN), Random Forest, and Support Vector Machine (SVM), using a heart disease dataset obtained from Kaggle. The dataset comprises 14 variables, including 13 attributes and 1 target variable. The models were tested on data split ratios of 70:30, 80:20, and 90:10, with performance measured by accuracy, precision, recall, and F1-score. The results showed that Random Forest performed best, achieving the highest accuracy of 98.54% at the 80:20 ratio. Decision Trees followed, yielding similar results, while K-NN performed the best at the 90:10 ratio, with a precision of 100%. In contrast, Naive Bayes performed lower due to high false positives. These results are in line with previous studies, confirming the effectiveness of Random Forest and Decision Trees in predicting heart disease risk. This research contributes to the development of reliable machine-learning models for early diagnosis and risk assessment of heart disease.

Mango Ripeness Classification Using Deep Learning-Based Convolutional Neural Network Models

2025-10-06T15:09:31+00:00

Manual assessment of mango ripeness is often subjective, time-consuming, and labor-intensive, making it inefficient for large-scale applications. This study investigates a deep learning–based approach for mango ripeness classification using Convolutional Neural Networks (CNN) and a publicly available image dataset obtained from Kaggle. Three CNN architectures, ResNet50V2, InceptionV3, and MobileNetV2, were evaluated using transfer learning and optimized with Adam and RMSprop to classify mangoes as unripe, ripe, or rotten. Data augmentation techniques were applied to improve model generalization and reduce overfitting. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results show that all models achieved competitive performance, with ResNet50V2 obtaining the highest test accuracy of 81.8%, while InceptionV3 demonstrated more stable validation results. These findings indicate that CNN-based transfer learning is effective for mango ripeness classification and can serve as a baseline reference for further development of automated fruit classification systems.

A Performance Comparison of Machine Learning Algorithms in Text Analysis of the Free Lunch Program

2025-10-06T15:03:40+00:00

The rise of public discourse on social policies in digital platforms necessitates accurate sentiment analysis to support data-driven policymaking. This study investigates the performance of three machine learning algorithms, Naive Bayes, Random Forest, and Support Vector Machine (SVM), in classifying public sentiment toward Indonesia’s Free Lunch Program. Data were collected via web scraping of YouTube comment sections, preprocessed with standard Natural Language Processing (NLP) techniques, and transformed with TF-IDF vectorization. Each algorithm was evaluated using accuracy, precision, recall, and F1-score on a three-class sentiment classification task with classes positive, neutral, and negative. Results show that Random Forest outperformed the other models with an accuracy of 93.55%, followed by SVM at 82.54%, and Naive Bayes at 75.71%. Random Forest demonstrated strong consistency across all sentiment categories, while SVM also delivered competitive results, especially for neutral sentiment. Naive Bayes, although efficient, struggled with imbalanced data, particularly in detecting negative sentiment. These findings highlight the superiority of ensemble models, such as Random Forest, in handling complex sentiment tasks and provide a foundation for developing robust NLP-based public opinion analysis tools for social policy evaluation. Unlike previous studies that primarily focused on product reviews or political sentiment, this research provides a comparative evaluation of machine learning algorithms for analyzing public opinion on Indonesia's Free Lunch Program using YouTube comments. The findings contribute practical guidance for selecting appropriate sentiment classification models to support evidence-based public policy evaluation

Exploring Deep Learning on Apple Stock Volatility: Long Short-Term Memory and Gated Recurrent Unit Performance

2026-07-13T11:27:12+00:00

Forecasting Apple Inc. (AAPL) stock prices is challenging due to the nonlinear and dynamic nature of financial data. This study compares the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures in predicting AAPL stock volatility. Daily stock data from January 1, 2020, to May 5, 2025, were used to evaluate both models under various hyperparameters. Performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results show that GRU with the AdamW optimizer (learning rate 0.01, batch size 8) achieved the best performance (MAE: 3.4188, RMSE: 4.9261, MAPE: 1.5648). Beyond identifying optimal hyperparameters, this study highlights critical architectural differences. The findings reveal that GRU's streamlined gating mechanism adapts more effectively and more rapidly to sudden price fluctuations than LSTM's complex memory-retention structure. These results demonstrate GRU's distinct advantage for computationally agile and accurate stock forecasting under highly volatile market conditions.

Comparative Analysis of Decision Tree and Random Forest Models for Student Exam Score Classification

2026-01-25T13:55:52+00:00

Advances in Educational Data Mining (EDM) have encouraged the use of machine learning algorithms to predict students’ academic performance. Although Decision Tree and Random Forest have been widely investigated, most previous studies compare them with multiple classifiers rather than conducting a focused evaluation under a unified framework. This study addresses this gap by systematically comparing the performance of Decision Tree and Random Forest in classifying student exam scores using the same preprocessing, cross-validation, and evaluation procedures. The dataset, obtained from Kaggle, consists of six attributes: student_id, hours studied, sleep hours, attendance percent, previous scores, and exam scores. The research methodology includes data collection, preprocessing, K-Fold Cross Validation, model training, and performance evaluation using a confusion matrix with accuracy, precision, and recall metrics. Experimental results show that Random Forest consistently outperforms Decision Tree, achieving a mean accuracy of 86.50%, compared with 81.50% for Decision Tree. In addition, Random Forest demonstrates greater stability and stronger generalization capability across different validation folds. These findings suggest that Random Forest is a more effective and reliable predictive model for supporting educational decision-making, particularly in identifying students at risk of poor academic performance and enabling timely academic interventions.

Comparison of Ensemble Learning Models for Hypertension Risk Prediction

2026-02-21T01:22:24+00:00

Hypertension remains a leading cause of cardiovascular disease worldwide, underscoring the need for accurate early risk prediction. This study aims to (1) compare the predictive performance of four representative machine learning models Decision Tree (DT), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and eXtreme Gradient Boosting (XGBoost) for hypertension risk prediction using a large, publicly available clinical dataset, and (2) examine the trade-off between predictive accuracy and model interpretability across these algorithms. The models were evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC-ROC), along with confusion-matrix-derived accuracy, precision, recall (sensitivity), specificity, and F1-score, under a k-fold cross-validation protocol. Results show that ensemble methods significantly outperform a single Decision Tree: LightGBM achieved the highest AUC (0.9976), followed by XGBoost (0.9966) and Random Forest (0.9923), while the Decision Tree attained an AUC of 0.9189. These results directly address the study objectives, demonstrating that gradient-boosting frameworks capture complex, non-linear relationships within medical data more effectively than a single interpretable tree, while Random Forest offers a favorable balance between accuracy and interpretability. The findings support integrating advanced ensemble models into clinical decision-support systems to enable more reliable, earlier hypertension risk stratification

A Comparative Analysis of Machine Learning Models for Weather Type Classification

2026-07-18T02:56:56+00:00

Accurate weather type classification plays an important role in supporting decision-making across various sectors, including transportation, agriculture, logistics, and disaster management. Although numerous studies have applied machine learning to weather prediction, most compare only a limited number of algorithms or evaluate them under a single experimental setting. This study addresses this gap by providing a comprehensive benchmark of five widely used classification algorithms Decision Tree, Random Forest, Support Vector Machine (SVM), Multilayer Perceptron (MLP), and Artificial Neural Network (ANN) using the publicly available Weather Type Classification dataset from Kaggle. These algorithms were selected because they represent interpretable, ensemble-based, kernel-based, and neural network approaches, enabling a comprehensive comparison of different machine learning paradigms. The dataset consists of structured meteorological attributes, including temperature, humidity, wind speed, precipitation, visibility, season, and location. Model performance was evaluated using three train-test split ratios (70/30, 80/20, and 90/10) and assessed through accuracy, precision, recall, confusion matrix, and ROC curve analyses. The experimental results show that ensemble-based and neural network models outperform single-tree approaches. Random Forest, MLP, and ANN demonstrate superior and more consistent performance across all data split ratios, with ROC-AUC values approaching 0.99. These findings establish a practical benchmark for selecting appropriate machine learning models for multi-class weather classification and demonstrate the effectiveness of ensemble and neural network approaches for developing reliable weather prediction systems using publicly available datasets.