Malware Detection: Performance Evaluation of ML Algorithms Based on Feature Selection and ANOVA
Nazma Akther, Md. Neamul Haque, Khaleque Md. Aashiq Kamal · 2023
67 Cy ber-attack is a critical problem in the field of digital world. To mitigate the cyber-attack, malware detection in the network traffic is the major concern. The success of an effective malware detection approaches on network traffic mainly depends on the correctness and building time of the approaches. Usually network traffic contains huge number of packets with vast features. In this paper, for minimizing the feature set we use four different feature selection techniques: CFS Subset eval, Consistency Subset eval, Infogain Attribute, One RA attribute etc. We have found that CFS Subset eval reduces the features of KDD dataset to 10 from 41 features (reduction rate is above 75%. Moreover, this work compares the following machine learning approaches: Naive Bayes, J48 as Decision Tree, Random Forest classifiers on the full and reduced featured data set. The result shows Random Forest classifier has the highest accuracy rate 99.97% with the One RA attribute selection technique. On the other hand, Naive Bayes has the lowest accuracy rate 90.62% with the Consistency feature selection technique. At final step, we have applied ANOVA technique to justify our experimental result statistically.