Optimized Zero False Positives Perceptron Training for Malware Detection

Abstract

The increasing number of malware in the past 4 years has determined researchers to test different machine learning techniques to automate the detection system. But because of the large size of the dataset and the need of having a high detection rate, the resulted models have often produced many false positives. This paper proposes a modified version of the perceptron algorithm able to detect malware samples while training at a low rate (even zero) of false positives. A very low number of false positives is crucial because in a real life situation detecting a clean file as malware can destroy the operating system or render other programs unusable. We also provide a method of optimizing the training speed for the algorithm while maintaining the same accuracy. The resulted algorithm can be used in an ensemble or voting system to increase detection and eliminate false positives.