Unsupervised anomaly detection based on a multi-layer perceptron

A method of unsupervised anomaly detection using a multi-layer perceptron was proposed to solve the problem that a mass of supervised data is needed to apply intrusion detection in computer systems. The network can realize functions of encoding and decoding. The main characteristics of the samples were learned under the principle of least mean square errors. The detailed learning algorithm was discussed. Tests indicate the feasibility of these algorithms. The method of unsupervised anomaly detection based on a multi-layer perceptron can detect intrusions without a mass of supervised data and is fit for application in intrusion detection systems.