Malware Detection with Different Voting Schemes
Jyoti Landage, Mayuri Wankhade · Zenodo (CERN European Organization for Nuclear Research) · 2014
A common way of launching the attack in computer system is Malware. It has malicious intent of performing any kind of malicious action to computer system as a result entire system crashes. It comes in different forms like virus, Trojan , Spyware, Scareware, Adware etc. Traditional malware detection techniques viz. signature-based, Heuristic-based and Specification-based detection technique are unable to detect some form of malware and each technique has its own advantages and disadvantages. A new methodology is proposed for malware detection that is based on data mining and machine learning techniques to detect known as well as unknown instances of malware. The new methodology uses disassemble process and three pre -processing techniques as part of feature extraction process to produce three different data sets with different configurations; feature selection technique is used to achieve consistent, reduced feature dataset. Three classification algorithms are used to generate and train the classifiers named as Ripper, C4.5 and IBk. The ensemble learning algorithm voting is used to improve the accuracy of result. Here majority voting and veto voting is used, the predicted output is decided on the basis of majority vo ting and veto voting. In veto voting the decision strategy of veto is improved by introducing the trust-based veto voting that definitely helps to improve the detection accuracy.