Multi-NetDroid: Multi-layer Perceptron Neural Network for Android Malware Detection

Abstract

Android malware detection has become a critical concern with the emergence of smartphones. Over the last few years, research has revealed a gradual improvement in the detection of malware from mobile operating systems through both static and dynamic analysis. Machine learning techniques are used to analyze various features and to train a larger dataset; to do this, a range of deep learning algorithms have been used previously. In this paper, we proposed a multi-layer perceptron (MLP) neural network for the Android malware detection method named as Multi-NetDroid. Training and evaluation of the proposed model have been done on publicly available datasets of the android applications. The data consist of features extracted from the manifest file (Intent, Permission), the dex file (API Call Signature), and command signatures existing within an APK file. The Multi-NetDroid model is built with four dense layers for training and classification of malware or benign applications and is evaluated as an improved classifier for Android malware detection. To evaluate our model performance we experimented with two separate datasets (Drebin-215 and Malgenome-15), and our model achieved 99.19% and 99.12% of accuracy. Furthermore, for the validation of our framework, we have also compared the results with different Machine Learning (ML) classifiers and found that our model Accuracy outperforms the classical ML methods.