Discovering Time-Series Building Blocks Using the ‘Significance Engine’

Abstract

This paper presents a time-series prediction framework called the “Significance Engine” which contributes to the area of time-series prediction by presenting a novel, noise resistant mechanism that combines together the techniques of Takens' Theorem and dynamic self-organising maps. It is built around the idea that time-series can be broken down into both random and non-random components and that the non-random components can be used to help predict future time-series movement. It is a highly parallel system which combines together multiple Dynamic Self-Organising Map units of the ‘Grow When Required’ type and time delay embedding techniques. It learns iteratively, so the more data the system is fed from the input distribution, the better the predictions it is able to make. After the framework has been initialised and ideally primed using historical data, it is able to recognise re-occurring patterns in highly noisy waveforms and to successfully make future predictions based on historically what occurred when these motifs were previously observed.