Time-based multi-layer perceptron for novelty detection in sensor networks

Abstract

This paper proposes a novelty detection system for sensor networks. The system utilizes a neural network function prediction methodology to predict sensor outputs in order to determine if the sensor outputs are novel. In addition to the novelty detection system, a modification to a standard neural network function predictor is proposed that allows the novelty detection system to quickly learn to accurately predict next sensor outputs. The parameter choice and the relationship between the threshold values and false alarm and missing attack rates are studied and recommendations are provided.