One side class SVM training methods for malware detection

Even though machine learning methods are being used in practice for malware detection, there are still many hurdles to overcome. Nowadays, there are still some challenges remaining regarding machine learning for malware detection: having a false positive rate as low as possible, fast classification, low volatile and disk memory usage. Because of these constraints, security solutions often have to rely on simpler models rather than on more complex ones. This paper has the purpose of reducing the training phase false positive rate of SVM models in the context of malware detection by using reformulations of the SVM optimization problem. The results obtained show that the proposed linear SVM model can be a drop in replacement with better false positive rate than regular linear SVM models or the one side class perceptron [4].