An Enhanced GHSOM for IDS

Artificial neural network, recently, is considered as a vibrant area in machine learning. Particularly, Growing Hierarchical Self Organizing Map (GHSOM) model, as an intelligent neural network, is vital in intrusion detection system (IDS). However, it suffers from prosaic topology, adheres to random weight vectors initialization, which degrades the performance metrics. In this paper, we progressively enhance the GHSOM to present a new delicate version, named EGHSOM. It consists of a meaningful initialization process instead of random initialization, a novel splitting threshold technique to stabilize the growth topology, merging and pruning methods on neurons to settle the topology and accelerate the detection, and a classification-confidence threshold to detect unknown anomaly in computer networks. The final model is trained using real-time traffic in addition to NSL-KDD and compared with other approaches. The result shows that EGHSOM is more efficacious than others and solves major drawbacks of intrusion detection in networks.