Random CapsNet forest model for imbalanced malware type classification task

Abstract Behavior of malware varies depending the types, which affects strategies system protection software. Many classification models, empowered by machine and/or deep learning, achieve superior accuracies for predicting types. Machine learning-based models need to do heavy feature engineering work, performance greatly. On other hand, require less effort in when compared that models. However, traditional learning architectures components, such as max and average pooling, cause architecture be more complex sensitive data. The capsule network architectures, on reduce aforementioned complexities eliminating pooling components. Additionally, based are data, unlike classical convolutional neural architectures. This paper proposes an ensemble model bootstrap aggregating technique. proposed method is tested two widely used, highly imbalanced datasets (Malimg BIG2015), the-state-of-the-art results well-known can used comparison purposes. achieves highest F-Score, 0.9820, BIG2015 dataset 0.9661, Malimg dataset. Our also reaches the-state-of-the-art, using 99.7% lower number trainable parameters than best literature.