Article Per Year (5 Year)
p-Index From 2021 - 2026
0.562
P-Index
This Author published in this journals
All Journal JOURNAL OF APPLIED INFORMATICS AND COMPUTING
Munirul Ula
Program Studi Magister Teknologi Informasi, Universitas Malikussaleh
Computer Science & IT

Published : 3 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 3 Documents
Search

 1 
Evaluation of Support Vector Machines and Adaptive Boosting in Classifying the Compliance Levels of Property and Building Taxpayers Using Receiver Operating Characteristic (ROC) Saumina Saumina; Munirul Ula; Asrianda Asrianda
Journal of Applied Informatics and Computing Vol. 10 No. 4 (2026): August 2026
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v10i4.13643

Abstract

Taxpayer compliance is a critical factor in increasing Property Tax (PBB) revenue. A low level of compliance can reduce local government revenue, making accurate classification methods essential for identifying taxpayer compliance. This study aims to compare the performance of Support Vector Machine (SVM) and Adaptive Boosting (AdaBoost) in classifying property taxpayer compliance. The dataset consisted of 58,998 property tax records collected from Lhokseumawe City, covering the districts of Banda Sakti, Blang Mangat, Muara Dua, and Muara Satu. The research stages included data preprocessing, label encoding, Min–Max normalization, data splitting using 80:20 and 70:30 scenarios, model training, and performance evaluation using accuracy, precision, recall, F1-score, and Area Under the Curve (AUC). Under the 80:20 data split, SVM achieved an accuracy of 92.89%, precision of 93.12%, recall of 98.81%, F1-score of 95.87%, and AUC of 80.59%, while AdaBoost achieved an accuracy of 92.86%, precision of 93.11%, recall of 98.78%, F1-score of 95.86%, and AUC of 80.76%. Under the 70:30 data split, SVM achieved an accuracy of 93.02%, precision of 93.18%, recall of 98.90%, F1-score of 95.95%, and AUC of 80.32%, whereas AdaBoost achieved an accuracy of 92.99%, precision of 93.18%, recall of 98.87%, F1-score of 95.93%, and AUC of 80.97%. Overall, both methods demonstrated comparable classification performance, while AdaBoost exhibited slightly better discriminative capability based on the AUC values.
Comparative Analysis of the SMART and ARAS Methods in a Decision Support System for Motorcycle Loan Applicant Eligibility Sri Kurnia; Dahlan Abdullah; Nurdin Nurdin; Munirul Ula; Muchlish Abdul Muthalib
Journal of Applied Informatics and Computing Vol. 10 No. 4 (2026): August 2026
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v10i4.13714

Abstract

Motorcycle financing is one of the most popular consumer financing products in Indonesia. However, the credit eligibility assessment process at PT Mega Central Finance (MCF) Lhokseumawe Branch is still performed manually, making it susceptible to inconsistent evaluations and increasing the risk of non-performing loans. This study aims to implement and compare two Multi-Criteria Decision-Making (MCDM) methods, namely the Simple Multi-Attribute Rating Technique (SMART) and the Additive Ratio Assessment (ARAS), for evaluating motorcycle financing eligibility based on seven criteria, including age, monthly income, occupation, marital status, number of dependents, outstanding debt balance, and down payment (DP). The study utilized primary data from 223 loan applicants collected during the 2024 to 2025 period through interviews and document reviews using a total sampling technique. Criterion weights were determined using expert judgment from credit analysts. The results show that the SMART method classified 96 applicants as eligible and 127 applicants as not eligible, whereas the ARAS method classified 181 applicants as eligible and 42 applicants as not eligible, using a minimum eligibility threshold of 0.60. The difference in the results is primarily attributed to the distinct normalization mechanisms of the two methods. SMART is more sensitive to extreme values in highly weighted criteria, resulting in a more selective evaluation process, whereas ARAS produces a more balanced distribution of preference scores by normalizing criterion values relative to the optimal solution, leading to a more flexible assessment. The findings indicate that the two methods complement each other. SMART is recommended for organizations adopting a conservative credit approval policy, while ARAS is more suitable for organizations seeking to expand the number of eligible applicants while maintaining a balanced consideration of all evaluation criteria.
Comparative Analysis of Random Forest and Long Short-Term Memory for Predicting Optical Power Degradation in FTTH Networks Hermansyah Hermansyah; Taufiq Taufiq; Defry Hamdhana; Munirul Ula; Muhammad Ikhwanus
Journal of Applied Informatics and Computing Vol. 10 No. 4 (2026): August 2026
Publisher : Politeknik Negeri Batam

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30871/jaic.v10i4.13717

Abstract

Fiber-to-the-Home (FTTH) networks are widely used to provide high-speed broadband services, but optical power degradation can reduce network performance and service quality. This study compares Random Forest (RF) and Long Short-Term Memory (LSTM) for predicting FTTH network conditions classified as Normal, Warning, and Critical. The study used 63,145 historical records collected from 58 Optical Network Terminals (ONTs) between March and May 2026. To provide a fair comparison, RF was trained using engineered tabular features, including lag and rolling-window statistics, while LSTM used six-step sequential data representing approximately the previous six hours. Model performance was evaluated using accuracy, precision, recall, F1-score, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), training time, and inference time. The results show that RF substantially outperformed LSTM, achieving 98.41% accuracy, precision, recall, and F1-score, with an MAE of 0.0173 and RMSE of 0.1420. RF also required only 2.667 seconds for training and 0.0102 ms for inference, compared with 189.98 seconds and 0.1856 ms for LSTM. Per-class evaluation confirmed that RF performed well across all network conditions, with precision and recall above 99% for Normal, above 94% for Warning, and above 90% for Critical. A strict chronological train-test split further confirmed the robustness of RF, which achieved 98.59% accuracy. Feature importance analysis showed that historical optical power, particularly lag-based features, was the most influential predictor of network degradation. These findings indicate that FTTH optical power degradation can be effectively modeled using engineered tabular features rather than a purely sequential approach. Finally, the RF model was integrated into a web-based monitoring dashboard with WhatsApp-based early warnings to support proactive FTTH network maintenance.
Co-Authors Asrianda Asrianda Dahlan Abdullah Defry Hamdhana Hermansyah Hermansyah Muchlish Abdul Muthalib Muhammad Ikhwanus Nurdin Nurdin Saumina Saumina Sri Kurnia Taufiq Taufiq