Effective Feature Extraction via N-Skip Gram Instruction Embedding Model using Deep Neural Network for designing Anti-Malware Application

Abstract

Due to increase in third party software release in open licensed platform, there is widespread on malware integration in such downloadable applications targeting individual or Corporate’s private data’s. On the other hand, due to variety of malware’s increase in market, it’s difficult to identify the new malware’s embedded in it. Alternatively, feature extraction and processing becomes very difficult for categorizing benign and malicious files altogether. By considering the above issues, this papers proposes an effective Instruc2vec framework for effective detection of malware’s dynamically. Existing works is demonstrated with static malware’s which lacks in identifying new types of malware’s. Initially, open licensed software’s are converted to instruction format via IDA Pro [19]. Extracting the features in instructions is a well-challenging task and feature representation also requires domain expert knowledge. This framework deploys Deep Neural network model naming convolutional neural networks to extract the operational opcodes from the instructional file. This framework extract the hidden patterns also in the file by dynamic learning process. Instruc2vec follows the word2vec process for feature selection and extraction procedures. In this experiment, real data collected from Linux popular packages and third party software downloads are used with different optimization and architecture to ensure the correctness of the application.