API-based features representation fusion for malware classification

This paper establishes a new feature representation fusion method to enhance the malware classification results. We designed a convolutional auto-encoder to learn several API-based (Application Programming Interface) malware representations with different latent dimensionalities. We used these representations to learn several neural network classifiers. We established a decision-level fusion technique to combine the individual decisions from each neural network classifier to obtain the final decision. We tested our method using a publicly available malware API dataset. We demonstrated that the proposed method improved the malware classification results.