A Hybrid Network Intrusion Detection Framework using Neural Network-Based Decision Tree Model

Abstract

Network Intrusion Detection System (NIDS) is a mechanism for detecting anomaly in computer networks. Several NIDS techniques have been developed in the past, but these techniques are still limited in detection accuracy, error rate and in detecting new attacks. In this study, a hybrid network intrusion detection framework using a neural network-based decision tree model for NIDS was developed. The developed model is divided into four modules: data collection, data preprocessing, feature selection and detection. The data collection module adapted the NSL-KDD dataset for implementation due to its modern attack representation. The data preprocessing module used the random undersampling technique to reduce the data imbalance problem. The feature selection module consists of a hybrid feature selection method to select the most important features from the adapted intrusion dataset. The detection module involves a neural network-based decision tree classifier for the automatic generation of rules for intrusion detection. The results showed that the developed model based on the full dataset is better than the other related methods with TP, FP, accuracy, precision, recall, and F1-score of 98.7, 1.3, 98.42%, 98.54%, 98.56% and 98.56% respectively. Similarly, the results showed that the developed method based on the reduced dataset is better than the other related methods with TP, FP, accuracy, precision, recall, and F1-score of 98.9, 1.2, 99.42%, 99.54%, 99.56%, and 99.56%, respectively.