A Hybrid Filter Based Classification Framework for Imbalance Cyber Intrusion Detection Databases

Abstract

Outlier detection, feature ranking and multi-class classification play a major role in the different real-time applications such as IoT databases, cyber-attack data, credit risk prediction and network intrusion detection. Most of the conventional multi-class prediction modes are developed for limited data size and features. Outlier prediction is one of the main issue in most of the numerical data types in cyber anomaly detection process. As the number of features and classes in the heterogeneous databases are increasing in size, traditional filter based ensemble learning approaches are difficult to improve the true positive rate and error rate. These traditional models are trained and tested with limited feature space without outlier detection. Also, it is difficult to improve the imbalance property of multiple classes due to the variation in data distribution and outliers. In this work, a hybrid outlier detection based multi-class classification framework is developed in order to improve the classification efficiency and error rate on heterogeneous datasets. Experimental results prove that the filter based multi-class classification framework has better optimization than the conventional filter based multi-class classification models.