NETWORK INTRUSION DETECTION USING RESILIENT BACKPROPAGATION

Abstract

Information is one of the most valuable possessions today. As the Internet expands both in number of hosts connected and number of services provided, security has become a key issue for the technology developers. With the growth of smart devices and internet technologies, anomalous traffic detection has become a major concern. This thesis is focused on the detection of attacks in a network by using Multilayer Perceptron (MLP) trained with Resilient Backpropagation. NSL-KDD dataset, an intrusion detection attacks database is used as an input dataset for network intrusion detection. In each record of the NSL-KDD dataset, there are 42 attributes, 41 attributes unfolding different features and a label assigned to each either as an attack type or as normal. The 2nd (Protocol_type), 3rd (Services), and 4th (Flag) attributes are converted into numerical format. The 42nd attribute is first classified under different category of attacks (DoS, Probe, R2L, U2R) or normal, then assigned a numerical value and finally, 5 bit code are assigned to each of them. Architecture of MLP is determined to have 41 neurons in the input layer, 30 neurons in the hidden layer, and 5 neurons in the output layer. The number of neurons in the hidden layer was fixed selecting that value for which Performance (MSE) was best. In this thesis, a Multilayer Perceptron is trained with Resilient Backpropagation algorithm and the research evaluates the performance of the algorithm. Out of the 47735 records, 40% of it is used for training neural network, 50% of it is used for validation, and 10% is used for testing. Different parameters (TP, FP, FN) are noted from the confusion matrix and recall rate, precision rate were calculated for each. The result showed overall recall rate and precision rate to be 99.8% and 99.83%. The overall detection rate of the system is found to be 96.7% which is better than any other existing backpropagation algorithms.