Malware Detection through Machine Learning Techniques

Malware attack is a never-ending cyber security issue. Since traditional approaches are less efficient in detecting newly appeared malware, researchers are applying machine learning methods. In this research we started by an overview of the domain and went over available malware datasets. Then we discussed disadvantages of traditional Anti-Malware methods and reviewed possible Machine Learning techniques used in this domain. A study on EMBER dataset has been made with an objective of improving the baseline Gradient Boosted Decision Tree model by optimizing its hyper-parameter and eliminating noisy features from the dataset. EMBER dataset consists of 1.1M observations of static features extracted from executable files. Our optimized model has achieved 99.38% accuracy with 0.004 false positive rate in 7 minutes running time. We conclude that Machine Learning techniques are practical to be applied as anti-malware solutions including for Zero-day attacks.