MPDGGA tops 10 of 11 intrusion detection datasets with minimal feature selection

Researchers have long struggled with feature selection for network intrusion detection systems (NIDS) due to the high dimensionality and redundancy of network traffic data. Now, Chunzhen Li's MPDGGA (Multi-population Diversity-guided Genetic Algorithm) offers a breakthrough solution. Built on a chained multi-population evolutionary structure, this algorithm introduces a diversity-guided operator that leverages information gain ratio to maintain population diversity and guide evolutionary operators more effectively than traditional methods. The approach directly addresses the common pitfalls of standard GA-based feature selectors: difficulty maintaining diversity and lack of operator guidance.

Li validated MPDGGA across 11 datasets including NSL-KDD, UNSW-NB15, and 9 UCI benchmark datasets. The results are compelling: MPDGGA achieved the highest classification accuracy on 10 of the 11 datasets, outperforming four other state-of-the-art multi-population feature selection models. Perhaps more impressively, it selected only 2.26% of the available features on average, dramatically reducing the feature space while improving or maintaining detection performance. This efficiency could translate to real-world NIDS deployments that are faster, more resource-efficient, and less prone to overfitting.