Malware Classification Using Multi-layer Perceptron Model

Malicious software harms computer and network resources. For malware detection, various machine learning and deep learning algorithms are used. In this paper, binary classification of benign and malware files is done using a multi-layer perceptron model using dynamic features. Both malware and benign samples are executed into the dynamic analysis environment using the Cuckoo sandbox. Cuckoo sandbox generates the detailed behavioural reports which consist of several runtime features like API calls, registry changes and network activities. These features are investigated and represented as a training feature set. After that multi-layer perceptron model is trained over the training feature set. Different parameter values of activation function, loss function and alpha (regularization parameter) are used for training the multi-layer perceptron model to develop the optimal malware classifier. The proposed malware classifier produced an accuracy of 99.2% using the Adam loss function in binary classification of the executable files.