A Hybrid Deep Convolution Neural Network Algorithm with Spatial Attention for Malware Detection in Android Operating System

The data security of organizations and individuals is largely threatened by the huge rise in malware in recent. Huge time and memory overheads are caused by the conventional static and dynamic defence and analysis techniques. An efficient deep learning model for malware detection is proposed in this work that can classify two-dimensional images that represent malware binaries using a visualization-based model. A hybrid deep convolution neural network model using spatial attention is used where the final classification layer is introduced with a reweighted class-balanced loss function. Four benchmark malware datasets are used for performing comprehensive experiments. From the experimental results of the proposed work, it is observed that the accuracy is higher compared with the conventional systems. Low computational time is maintained, while the false positive rates are considerably reduced when compared with the conventional malware mitigation models. Obfuscation attacks are also effectively mitigated using the proposed technique.