Claim Missing Document
Check
Articles

Found 2 Documents
Search
Journal : Infotech: Journal of Technology Information

ANALISIS KINERJA ALGORITMA MACHINE LEARNING DALAM MENDETEKSI ANOMALI KETINGGIAN AIR LAUT: STUDI PERBANDINGAN ONE-CLASS SVM DAN ISOLATION FOREST Alifandra, Dhafa; Pratiwi, Nunik
Infotech: Journal of Technology Information Vol 11, No 2 (2025): NOVEMBER
Publisher : ISTEK WIDURI

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.37365/jti.v11i2.405

Abstract

This study aims to compare the performance of two machine learning algorithms for anomaly detection One-Class SVM and Isolation Forest in identifying anomalies in sea level data in Indonesia, a region with high tsunami risk. The data were obtained from an official Indonesian government source over a one-year period and underwent preprocessing, including data cleaning and standardization. The models were evaluated using statistical analysis (Mann-Whitney U test), clustering metrics (Davies-Bouldin Index and Silhouette Score), and visual inspection. The results indicate that Isolation Forest outperformed the other algorithm with a Davies-Bouldin Index of 0.8124, while One-Class SVM achieved the highest Silhouette Score at 0.4381, although its Davies-Bouldin Index was higher at 0.9163. This study contributes to the selection of effective algorithms for ocean monitoring systems as part of disaster mitigation strategies in Indonesia.
PREDIKSI HARGA DAN RISIKO SAHAM TELKOM DAN INDOSAT MENGGUNAKAN LSTM DAN VAR DENGAN VISUALISASI Alam, Indera Nurul; Pratiwi, Nunik
Infotech: Journal of Technology Information Vol 11, No 2 (2025): NOVEMBER
Publisher : ISTEK WIDURI

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.37365/jti.v11i2.399

Abstract

Investment in Indonesia shows a growing trend, with Telkom (TLKM) and Indosat (ISAT) being among the mostactively traded stocks with high volatility. This condition raises the need for reliable stock price prediction andinvestment risk analysis. This study aims to develop a daily stock closing price prediction model using the Long Short-Term Memory (LSTM) algorithm with a Bidirectional LSTM architecture and to conduct risk analysis based on Valueat Risk (VaR) through parametric Monte Carlo simulation. Fourteen years of historical stock data were utilized andprocessed through feature engineering techniques (return, moving average, volatility) and 30-day windowing.Baseline models with one to four layers were tested, and the best model was further optimized through hyperparametertuning using the Random Search method. The results indicate that the single-layer Bidirectional LSTM modeldemonstrated the best performance on the testing data. Evaluation shows a significant performance improvement aftertuning, with RMSE decreasing from 69 to 67, MAPE from 1.61% to 1.59%, and R-Square remaining high at 0.97 forTelkom, as well as a reduction in RMSE from 91 to 74, MAPE from 2.81% to 2.27%, and an increase in R-Squarefrom 0.76 to 0.84 for Indosat. The VaR analysis reveals that the predicted daily and 80-day risk values show onlyminor deviations from the actual values, supporting the validity