Optimized Deep Learning for Android Malware Detection A PSO-LSTM Hybrid Approach

Abstract

Introduction: The widespread adoption of Android as the most used mobile operating system has made it a prime target of malicious applications that exploit its permission-based access control system. While this access system helps to protect user data, it is misused by malware to gain unauthorized access to sensitive information. As a result, getting an efficient Android malware detection has become a critical area of research.

Objectives: This paper addresses the problem by exploring multiple machine learning approaches to detect malware based on permission. Furthermore, to enhance the detection performance and model robustness, we propose a hybrid approach that combines PSO and LSTM neural networks.

Methods: In our hybrid approach, PSO is employed to automatically optimize the hyperparameters of the LSTM model. We evaluate our approach using three publicly available Android malware datasets. The first is the Malware/Benign Permissions Android dataset (C.Urcuqui and A.Navarro 2016, U.Christian 2016), the second is the Drebin dataset (Soju n.d.), and the third is the Droid det malware dataset (F. Akbar 2025).

Results: The experimental results demonstrate that the hybrid model PSO-LSTM outperforms traditional machine learning techniques in terms of detection performance. We also compare our results with other existing studies in the field, showing that our method offers a more effective and scalable solution for Android malware detection.

Conclusions: The PSO-LSTM hybrid model enhances the detection of Android malware by the optimization power of PSO in the selection of hyperparameters of the LSTM. This method offers a scalable solution for malware detection, providing an effective approach that surpasses existing models in the field.