A Novel Detection Technique using Martingales For Optimal Parameters in Time Series

Time series analysis is becoming essential in different areas for the observation and monitoring of time-sequential data sets to extract relevant statistics and predict the series’ behaviour. Current approaches effectively detect changes in the data streams. Still, most of these techniques are limited to noise interference and the inability to identify the most significant parameter values for productive abnormality detection in time series. In this paper, we improve on the previous moving median of the martingale sequence and the Gaussian moving average of the martingale sequence approaches by implementing various optimisation algorithms such as G-mean enumeration, genetic algorithms and particle swarm optimisation. The use of these methods allows us to find the optimal parameter set for each algorithm. The proposed system can reduce noise in the data and estimate the change degree in time series scenarios. Results show that the proposed approaches perform better than the previous martingale approaches.