On the Capability of SOINN Based Intrusion Detection Systems
The Self-Organizing Incremental Neural Network (SOINN) features online unsupervised learning, which is attractive for building an intrusion detection system that is accurate and space efficient. This paper explores the detection capability of improved single layer SOINN based methods, including unsupervised and semi-supervised methods. Experiments are carried out on the KDD Cup 99 data set. The results show that in the scenario where expert labeling is not present, unsupervised detection method based on SOINN outperforms existing methods in detection rate (0.8 percent improvement) and false alarm rate (2.4 percent improvement).