Feature Weighting and Selection for a Real-Time Network Intrusion Detection System Based on GA withKNN

摘要

A good feature selection policy which can choose significant and as less as possible features plays a key role for any successful NIDS. The paper presents a genetic algorithm combined with kNN (k-Nearest Neighbor) for feature weighting. We weight all initial 35 features in the training phase and then select tops of them to implement a NIDS for testing. Many DoS/DDoS attacks are applied to evaluate the system. For known attacks we can get the best 97.42% overall accuracy rate while only the top 19 features are considered; as for unknown attacks, we can get the best 78% overall accuracy rate by top 28 features.