A HYBRID DNN-DKN MODEL WITH MAPREDUCE FRAMEWORK FOR DETECTING MALICIOUS THREATS IN LARGE EXECUTABLE FILES

In the current world, Malware has become a serious hazard to network security. Malware is considered as malicious software and detection of malicious threats in large executable files is a crucial aspect of improving cybersecurity. Malicious detection in large executable files is complex due to the complexity of threats, and increasing volume of data. Hence, the MapReduce framework is utilized to analyze large datasets effectively. Therefore, this paper presents the Hybrid Deep Neural Network-Deep Kronecker Network (DNN-DKN) model for detecting malicious threats using the MapReduce framework. Initially, input executable files are forwarded to the data partitioning, where Fuzzy Local Information C-Means (FLICM) is employed for data partitioning. Then, partitioned files are sent to the Map Reduce framework, which encompasses the mapper and reducer phases. In the mapper phase, the features are extracted from the partitioned file using Binary hexadecimal and Dynamic Link Library (DLL). Then, extracted features are augmented by the oversampling method. Subsequently, the augmented files from the mappers are combined and passed to the reducer phase, where the malicious threats are detected using the proposed DNN-DKN. The DNN-DKN is the combination of a Deep Neural Network (DNN) and a Deep Kronecker Network (DKN). The DNN-DKN outperformed the existing methods and established performance metrics with minimum False Positive Rate (FPR), maximum True Positive Rate (TPR), and, maximum accuracy of 9.10%, 94.79% and 91.66%.