Machine Learning Empowered Efficient Intrusion Detection Framework

Abstract

In modern era security is becoming major and basic need of any system. Protecting of a system from unauthorized access is very important for a network system. Network security is turning out to be an influential subject in information technology territory. Hackers and squatters commit uncountable successful attempts to intrude into networks. Intrusion Detection System plays a vital role in a network security to identify and detect the anomalies in a security system of network. The performance of IDS can be measured through its intelligence, efficiency and accurate detection of unknown and known attacks. The greater the gain concept give the best possible detection rate of anomalies. This study proposed a machine learning framework based on MLP classifier with accuracy 99.98%. This work is further validated through 10-fold and JackKnife cross validation. Key metrics to see the impact on accuracy and other performance measured metrics such as Sensitivity, Specificity and Matthew’s Correlation Coefficient. All the metrics gained their highest ratio, which means MLP is the best classification technique. The accuracy, sensitivity, specificity and MCC rate of the suggested model computed 99.99% from whole dataset of UNSW-NB15. These results show the improvement in accuracy while applying different perceptron topologies. K-fold and JackKnife topologies are capable to earn the 99.99% accuracy