A Deep Learning Approach for Static Analysis-Based Android Malware Detection
Author : Dr. S Krishnaveni and Reshmitha D
Abstract :
This project presents a deep learning-based Android malware detection system that leverages multiple neural network architectures—Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and 1D Residual Network (ResNet)—to classify applications as malicious or benign based on extracted static features. The dataset is preprocessed using standard normalization techniques and split for model training and evaluation. Each model is assessed using accuracy and confusion matrix metrics, with ResNet achieving the highest accuracy among all.
To enhance usability, the CNN model is integrated into a Flask-based web application equipped with a desktop-style graphical user interface using PyWebView. This GUI allows users to upload feature datasets in CSV format and receive real-time malware predictions in an interactive environment. The system is further improved with reproducibility settings, learning rate tuning, and early stopping callbacks to ensure model stability. The combination of robust model performance and an intuitive user interface demonstrates the system’s potential for practical and scalable Android malware detection.
Keywords :
Deep learning, Convolutional neural networks (CNN), Long short-term memory (LSTM), Residual neural networks (ResNet), Multilayer perceptron (MLP), Web-based application, Mobile application security, Graphical user interface (GUI).
