Optimized Machine Learning for Cyber Security Applications

Abstract

Machine learning plays crucial role in cyber-attack detection from different platforms like IOT (Internet of things), from network or from different ecommerce as well as banking sector. In this study different traditional machine learning and optimized machine learning algorithms are used for performance analysis. It is observed that optimized machine learning algorithms are performing better than traditional machine learning classifiers giving better performance metrics. The dataset of BOTNET attackers is used which has two labels as ‘Normal’ or ‘Attack’. The dataset is highly imbalanced so to avoid the training issues from highly imbalanced dataset in proposed method SMOTE algorithm is used which makes both normal user and attackers data equally available for training. The classifiers used in this proposed method are SVM (support vector machine) classifier , decision tree classifier , BOGP Optimized Decision tree and advance deep learning algorithm CNN (Convolution Neural Networks). Out of these four classifiers CNN is performing superior than other traditional classifiers.