IJATIS: Indonesian Journal of Applied Technology and Innovation Science

PM 2.5 Prediction Using the Long Short-Term Memory Algorithm

Syaid El Hasyim — 2025-09-04

Air pollution poses a serious threat to human health and the environment, with far-reaching impacts on various aspects of life. Among its most harmful components is particulate matter less than 2.5 micrometers in diameter (PM2.5), which contributes significantly to degraded air quality. Accurate prediction of PM2.5 concentrations is crucial for public health protection and policy-making. This study employs the Long Short-Term Memory (LSTM) algorithm, a deep learning method well-suited for modeling large, complex, and time-dependent datasets, to forecast PM2.5 levels in Delhi, India. The dataset comprises daily records from January 1, 2015, to July 1, 2020. The proposed model achieved a Mean Absolute Percentage Error (MAPE) of 25.22%, indicating moderate predictive accuracy. These results demonstrate that the LSTM algorithm can serve as an effective tool for forecasting PM2.5 concentrations, providing valuable insights for air quality management and environmental planning.

Comparison of MAUT and EDAS Methods for News Media Selection on Youtube Platform Using ROC Weighting

Rizki Aulia Putra — 2025-09-04

News media is one of the things that is considered in getting information. In order to avoid issues or news that are not good, especially those smelling of politics in Indonesia, it is necessary to choose which media can be used as a reference, especially on YouTube as an alternative place to watch and get news. This problem can be solved with the help of a decision support system, which can choose an alternative news media with the help of the MAUT and EDAS methods. The alternatives tested for calculation are obtained based on how often the media is trending on YouTube. In future calculations, based on criteria determined from the results of expert interviews, namely transparency, credibility, news sources and subscribers will be tested to get the best news media with the help of ROC weighting. With the criteria that have been determined and the calculation of the MAUT and EDAS methods, getting the best news media ranking results from both methods is CNN Indonesia with a final value of preference 1. These results will be an illustration for the people of Indonesia in sorting out the news on each news media on YouTube.

Implementation of Machine Learning Algorithm for Heart Attack Disease Prediction

Febbi Ardiani — 2025-09-04

Heart attack disease is one of the leading causes of death worldwide, making early detection a critical factor in reducing mortality. However, manual prediction is often inaccurate due to the complexity of medical data. To address this issue, this study evaluates five machine learning algorithms K-Nearest Neighbors (KNN), Decision Tree, Naïve Bayes, Random Forest, and Support Vector Machine (SVM) for predicting heart attack risk. The dataset, obtained from Kaggle, was preprocessed and divided into training and testing sets using 70:30 and 80:20 ratios. Algorithm performance was assessed using accuracy, precision, recall, and F1-score. The results showed that Decision Tree and Random Forest achieved the best performance with accuracy up to 97.98%, while KNN recorded the lowest accuracy at around 61.36%. This study not only demonstrates the comparative effectiveness of these algorithms on the same dataset, contributing to the growing body of research on AI in healthcare, but also highlights their potential clinical utility. In particular, Decision Tree and Random Forest can support the development of AI-based clinical decision support systems to assist healthcare professionals in early diagnosis and risk management

Comparison of Supervised Learning Algorithms for Cancer Prediction

Intan Adha Maharani — 2025-09-04

This study focuses on the application of Machine Learning algorithms for cancer prediction using a classification dataset. Several algorithms were employed, including K-Nearest Neighbor (KNN), Naive Bayes Classifier, Decision Tree, Random Forest, and Support Vector Machine (SVM). The primary goal of this research is to evaluate the performance of each algorithm to identify the best method for achieving high accuracy in cancer classification prediction. The experimental results reveal variations in performance among these algorithms. The evaluation was conducted using metrics such as accuracy, precision, recall, and F1-Score. Based on the analysis, Random Forest and Support Vector Machine demonstrated the best performance with the highest accuracy compared to other algorithms. Meanwhile, the Naive Bayes algorithm tended to exhibit lower performance in predictions. This study emphasizes the importance of selecting the appropriate algorithm in the implementation of Machine Learning for medical applications such as cancer prediction. With these findings, it is hoped that the identified methods can assist in clinical decision-making and improve the accuracy of early cancer diagnosis.

Clasification of A Credit Card Fraud Detection Model Using XGBoost with Smote and Gridsearchcv Optimization

Amelia Rahmadani — 2025-09-04

The development of digital technology has motivated rapid growth in online transactions, so the increase in the volume of digital transactions also increases the risk of credit card fraud, particularly in transactions where a card is not present. By employing the Extreme Gradient Boosting (XGBoost) method in conjunction with the Synthetic Minority Over-sampling Technique (SMOTE) to solve class imbalance and fine-tuning model parameters using GridSearchCV, this study aims to improve a fraud detection system. The dataset, which consists of anonymized credit card transactions, presents a stark imbalance with fraudulent cases accounting for only 0.172% of the data. The study involves several stages: preprocessing the data, balancing class distribution, training the model, and evaluating its performance through metrics such as F1-score, precision, recall, accuracy, and AUC-ROC. Implementation of SMOTE proved effective in enhancing the representation of rare fraud cases without introducing overfitting, while GridSearchCV identified the most effective parameter configuration. The resulting model achieved top-tier performance with 100% accuracy, 0.81 precision, 0.85 recall, an F1-score of 0.83, and an AUC-ROC of 0.979, indicating strong capability in distinguishing fraudulent from legitimate transactions. The novelty of this study lies in the systematic integration of SMOTE, XGBoost, and GridSearchCV into a unified pipeline designed to address extreme class imbalance in real-world credit card transactions. Unlike previous studies that focused solely on algorithm comparison or hyperparameter tuning, this research emphasizes reducing false negatives, which pose the greatest financial and reputational risks. The findings not only demonstrate superior performance metrics but also provide practical contributions for financial institutions, regulators, and e-commerce platforms in developing scalable, reliable, and adaptive fraud detection systems

Comparison of ResNet50V2 and InceptionV3 with Adam, SGD, RMSprop Optimizers for Road Image Classification

Rifka Anrahvi — 2025-09-09

This study compares two Convolutional Neural Network (CNN) architectures ResNet50V2 and InceptionV3 with three optimizers (Adam, RMSprop, and SGD) for road condition classification. Using a dataset of 1,000 images categorized into four classes, the models were evaluated based on accuracy, precision, recall, and F1-score. Based on the results, ResNet50V2 with Adam optimizer performed the best, achieving 99% accuracy, whereas SGD yielded less-than-ideal results. This study is interesting since it compares architecture–optimizer pairings, a topic that hasn't been extensively studied in other studies. The results offer useful information for creating automated and dependable road monitoring systems that facilitate effective infrastructure upkeep. To further enhance performance, future study might entail implementing regularization techniques and growing the dataset.

System Usability Scale in Information System Application Development Using Systematic Mapping Study

Rahma Yani — 2025-09-09

The System Usability Scale (SUS) is a widely used usability evaluation method due to its simplicity and reliability. However, no study has systematically mapped the application of SUS in information system applications. This research aims to conduct a Systematic Mapping Study (SMS) to analyze the use of SUS in 30 international journals indexed in SpringerLink during the 2021–2025 period, focusing on publication trends, application domains, number and types of respondents, respondent criteria, and SUS score results. After the screening process, 28 relevant articles were identified, of which 14 directly employed SUS. The mapping results indicate that the technology domain dominates the application of SUS, with most respondents being general application users, while publication trends show fluctuations with a peak in 2025. These findings are consistent with the research objective, namely to provide a comprehensive overview of SUS usage patterns. The novelty of this study lies in mapping respondent characteristics and variations in SUS scores within the context of information system applications, which has not been systematically mapped before. The results are expected to serve as both an academic reference and practical guidance for developers and researchers in improving system design based on user experience.