Anomaly Detection in Time Series Data Using Spiking Neural Network

Abstract

One of the crucial issues in anomaly detection problems is identifying abnormal patterns in time series data that contains noise and in unstructured form. In order to deal with this problem, a good detector is needed with a capability to learn the complex features in the datasets and extract useful information to distinguish normal and abnormal patterns in the datasets. This study exploits the features of Spiking Neural Network (SNN) to generate potential neurons through its learning. These neurons will spike whenever it detects abnormal pattern in the data. The proposed method is consisting of three stages: 1) initializing the weight values using rank order method; 2) representing the real input data into spike values using Gaussian Receptive Fields; and 3) identifying the firing nodes that indicate the abnormal data. We applied the proposed technique to selected data with anomalies from time series datasets. Experimental results show that the proposed technique is capable of detecting the anomalies in the datasets with reasonable False Alarm Rate.