Enhanced Malware Image Classification through Ensemble model

Classification of malwares and viruses is a very important work in the cyber security field to protect the computers and systems from threats and attacks. In this paper, we proposed a novel approach for the classification of the malwares by combining 2 models EfficientNet B0 and EffiCBNet models. Here we have used Malimg dataset which is available in Kaggle and it contains around 9,339 malware images and these malwares are from 25 different families, for evaluating the models. EffcientNetB0 is a Convolution Neural Network (CNN) model, which showed great and high results on various image classification works. We used transfer learning to fine tune the EfficientNet model; whereas the proposed EffiCBNet model is a deep learning model which contains of convolution layers and batch normalization layers. The results have shown that the proposed model has given the highest accuracy of 99.70 on the dataset we have used, exceeding all other models. The proposed model is the combination of EfficientNet and EffiCBNet, which showed great results for the classification. The proposed model can also be used in real time applications for detecting malware attacks and threats