JTIULM (Jurnal Teknologi Informasi Universitas Lambung Mangkurat)
Vol. 11 No. 2 (2026)

Implementation of Autoencoder Model Based on Deep Learning for Keylogger Activity Detection on Android Through System Anomaly Analysis

Naufal Fauzi Akbar (Universitas Pamulang)
Saprudin (Universitas Pamulang)



Article Info

Publish Date
31 Oct 2026

Abstract

The increasingly widespread use of Android devices has heightened the risk to personal data security, particularly through keylogger activities capable of covertly intercepting user keystrokes without being detected by conventional signature-based security systems. This study aims to design and implement a keylogger activity detection system on Android devices using a deep learning-based Autoencoder model through a system anomaly analysis approach. A quantitative experimental method with an unsupervised anomaly detection approach was employed, in which the model was trained exclusively on normal device condition data. Data were collected through direct observation and internal logging mechanisms implemented in the developed KeyGuard application, yielding 904 data samples comprising CPU usage, RAM usage, network traffic, Accessibility Status, and Accessibility Events parameters, consisting of 764 normal-condition samples and 140 anomaly-condition samples obtained through testing with the Tispy keylogger application. The dataset was partitioned into training and testing sets, with evaluation conducted on 292 testing samples using a Confusion Matrix. The results indicate that the Autoencoder model, with a threshold value of 0.36, achieved an Accuracy of 96.57%, Precision of 95.77%, Recall of 97.14%, and F1-Score of 96.45%. The system successfully detected 136 of 140 tested keylogger activities, with a low false positive rate of six samples. This study concludes that the Autoencoder-based system anomaly analysis approach is effective for detecting keylogger activity on Android devices without relying on a signature database, thereby contributing to the development of more adaptive mobile security solutions capable of identifying previously unrecognized threats.

Copyrights © 2026






Journal Info

Abbrev

jtiulm

Publisher

Subject

Computer Science & IT Electrical & Electronics Engineering Engineering

Description

Jurnal Teknologi Informasi Universitas Lambung Mangkurat (JTIULM) is intended as a media for scientific studies on the results of research, thinking and analytical-critical studies regarding research in Systems Engineering, Informatics / Information Technology, Information Management and Information ...