Publication

Abstract : Domain generation algorithm (DGA) is the foundation of malware families because of the very fact that DGA generates huge variety of pseudorandom domain names to associate to a command and control (C2C) infrastructures. This paper focuses in handling classification of class in a imbalanced data. Cost-sensitive long short-term memory (CS-LSTM) approach is proposed which helps in understanding the importance of each class. Detecting malicious domain names (DMD 2018) data set is used. Optimal parameters are set to deep learning architectures using hyper-parameter approach. Experiments on CS-LSTM performed well compared to other deep learning architectures. Using this approach, 74.3% accuracy is obtained, varies 4.6% from the top scored system in DMD-2018.