Journal of Data Processing Leelavathi B, Rajesh Babu M 8 2 2018 10.6025/jdp/2018/8/2/63-73 http://www.dline.info/jdp/fulltext/v8n2/jdpv8n2_2.pdf Nowadays, detecting unknown computer worms is a challenging task. Anti-virus is the feasible solution for this problem but anti viruses that contains signature based methods are clueless against new unknown worms and it cannot detect new worms until it is updated with new worm signatures. This paper focuses on improving the accuracy of detecting the unknown worm while reducing the feature set extracted using DFA. Gain ratio method is implemented to reduce the amount of features. Keygraph and data crystallization classifier is proposed for classifying the unknown attacks. The dataset is collected from the VX Heaven virus collection website which contains the sources and samples of the virus collection. Extensive experiment was performed for testing the feasibility of detecting unknown worms. The evaluation results shows that by applying classification algorithms, the detection ratio obtained for specific unknown worms exceeds 98%.