IJATIS: Indonesian Journal of Applied Technology and Innovation Science

Implementation of Machine Learning Algorithms for Predicting Student Academic Performance

Amelianti Aman — 2026-03-17

This study examines the effectiveness of five data mining algorithms, K-Nearest Neighbor (K-NN), Naive Bayes, Decision Tree, Random Forest, and Support Vector Machine (SVM), in predicting and comparing students' academic performance. The goal is to evaluate the following: the study data includes average grades, learning motivation, study hours per week, and parental support. The data underwent preprocessing steps, including normalization, outlier removal, and splitting into training and test sets. Model performance was evaluated using accuracy, precision, and recall metrics. The results indicate that the Random Forest algorithm performed the best, followed by the Decision Tree, which also demonstrated strong performance. The SVM and Naive Bayes algorithms provided excellent results, while K-NN performed poorly due to class overlap in the data. The conclusion of this study is that the Random Forest algorithm is the most effective method for predicting students' academic performance and significantly contributes to data-driven analysis to improve the quality of education.

Comparison of Deep Neural Network and Convolutional Neural Network Algorithms for Bone Fracture

Ahmeid Aqeil — 2026-03-17

Bone fracture is a common medical condition that often affects elderly populations or individuals with degenerative diseases such as osteoporosis. Manual classification of fractures from X-ray images presents diagnostic challenges due to visual complexity and interobserver variability. In this study, we implemented and compared Deep Neural Network (DNN) and Convolutional Neural Network (CNN) architectures to classify bone fractures from radiographic images. The dataset consisted of 4099 X-ray images divided into fractured and non-fractured categories. Each model was trained using preprocessed and augmented data and evaluated using accuracy, precision, recall, and F1-score metrics. The evaluation results showed that the CNN model achieved better classification performance, with an accuracy of 80% and balanced class scores. In contrast, the DNN model showed poor generalization and strong bias toward the fractured class, yielding only 51% accuracy. This study concludes that CNN are more suitable for bone fracture classification tasks due to their superior ability to extract spatial features and generalize across categories.

Prediction of Fetal Health Using Machine Learning Algorithms

Dinda Mustika — 2026-03-17

This study evaluates several machine learning algorithms for predicting fetal health conditions using cardiotocography (CTG) data. The dataset contains 2,126 records with 22 numerical features obtained from Kaggle and is classified into three categories: normal, suspect, and pathological. Four classification models Decision Tree, Random Forest, Support Vector Machine, and Logistic Regression were implemented and evaluated using two data split scenarios (80:20 and 70:30). Model performance was assessed using precision, recall, and F1-score. The results show that Random Forest achieves the best performance with an F1-score of 91% in both split scenarios, indicating stable and accurate classification compared with other models. The contribution of this study is to provide a comparative evaluation of classical machine learning algorithms for CTG-based fetal health prediction. The findings can support the development of decision-support tools to help medical personnel detect and monitor fetal health risks early.

Comparison of Machine Learning Algorithm Performance for Toddler Stunting Prediction

Sophia Anjani — 2026-03-17

Stunting is a chronic nutritional issue in toddlers that has long-term effects on children's physical growth and cognitive development. This study aims to compare the performance of four machine learning algorithms, namely Support Vector Machine (SVM), Random Forest (RF), Multilayer Perceptron (MLP), and Logistic Regression (LR), in classifying the nutritional status of toddlers. The research stages included data preprocessing, data division into training and test sets, model training, and evaluation using accuracy, precision, recall, F1-Score, a confusion matrix, and Area Under the Curve (AUC). The evaluation results showed that Random Forest achieved the best performance, with an accuracy of 94%, as well as precision, recall, and F1-score values above 90%, and an AUC value close to 1.00 across all nutritional status classes. This was followed by the MLP algorithm in second place, with an accuracy of 93.29%. The main contribution of this study is the identification of a high-performing, stable model for large-scale stunting detection, providing a strong foundation for developing decision-support systems for early detection in the public health sector.

Classification-Based Supervised Learning Algorithms for Accurate Prediction of Customer Churn in Banking

Nora Waningsih — 2026-03-17

The banking industry has become increasingly dynamic with the emergence of financial technology (fintech) companies that have significantly changed customer behavior and expectations. As competition intensifies, customer churn has become a critical issue because it directly affects a bank’s revenue, reputation, and long-term sustainability. Therefore, banks require effective analytical approaches to identify customers likely to leave and to develop appropriate retention strategies. This study aims to analyze and predict customer churn likelihood using a bank customer dataset by applying supervised machine learning classification techniques. Five algorithms were evaluated, namely Decision Tree, Random Forest, Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The models were trained and evaluated using a hold-out validation approach, and performance was assessed using accuracy as the primary evaluation metric. The experimental results show that Random Forest achieved the highest accuracy of 86%, outperforming the other algorithms, while the MLP model produced the lowest accuracy of 82%. These findings indicate that ensemble-based methods provide better performance for predicting bank customer churn. The results of this study can assist banks in identifying potential churn customers and in developing effective customer retention strategies. Future research may explore additional algorithms, advanced data preprocessing techniques, and larger datasets to further improve prediction performance.

Comparative Analysis of Machine Learning Algorithms for Predicting Heart Attack

Habib Ahmad Tsaqif — 2026-03-18

Early detection of heart attack risk is crucial for reducing mortality rates associated with cardiovascular diseases. This study aims to perform a comparative performance analysis of four machine learning algorithms: Decision Tree, Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) in classifying heart attack risk using a clinical dataset from Kaggle. The research methodology includes data preprocessing, data splitting using a 70:30 hold-out scheme, and model evaluation through a confusion matrix and standard classification metrics. The test results indicate that Random Forest provides the superior performance with the highest accuracy of 84%. Meanwhile, the SVM and XGBoost algorithms achieved 80% accuracy each, while the Decision Tree achieved the lowest at 70%. These findings confirm that ensemble-based models like Random Forests exhibit greater stability in handling complex clinical data patterns, making them highly promising for integration into early heart health warning systems.

Performance Comparison of Five Machine Learning Algorithms for Early Detection of Alzheimer's Disease

Elsya Avivi — 2026-03-24

Alzheimer's disease (AD) is the most common form of dementia and is characterized by progressive cognitive decline. Early detection of AD is crucial for earlier intervention, as there is currently no cure for this disease. This study evaluates the performance of five machine learning algorithms, namely Logistic Regression, Decision Tree, Support Vector Machine (SVM), Random Forest, and XGBoost for AD classification using a dataset of demographic information, lifestyle, medical factors, and cognitive symptoms of patients. The data was processed through pre-processing steps (data cleaning, missing value imputation, and feature selection) and model evaluation using k-fold cross-validation with a 70:30, 80:20, and 90:10 data split. Unlike several previous studies that only conducted partial evaluations, this study directly tested the performance (head-to-head) of five algorithms representing various classification paradigms.The model evaluation also focused on maximizing Recall (Sensitivity) to minimize the critical risk of false negative diagnoses in the early detection process. The results showed that the XGBoost algorithm performed best across all evaluation metrics. With an 80:20 data split, XGBoost achieved the highest performance with Accuracy, Precision, and Recall of 95.1%. These findings demonstrate the effectiveness of the XGBoost algorithm in classifying patients and support the development of faster and more objective medical decision support systems. These results have practical implications that the ML model has the potential to support clinical decision support systems for the early detection of Alzheimer's disease