Timeweaver: a genetic algorithm for identifying predictive patterns in sequences of events

Abstract

Learning to predict future events from sequences of past events is an important, real-world, problem that arises in many contexts. This paper describes Timeweaver, a genetic-based machine learning system that solves the event prediction problem by identifying predictive temporal and sequential patterns within data. Timeweaver is applied to the task of learning to predict telecommunication equipment failures from 250,000 alarm messages and is shown to outperform existing methods. 1 INTRODUCTION Data is being generated and stored at an ever-increasing pace, and, partly as a consequence of this, there has been increased interest in how machine learning and statistical techniques can be employed to extract useful knowledge from this data. When this data is time-series data, it is often important to be able to predict future behavior based on past data. In this paper we are interested in the problem of predicting specific types of rare future events, which we refer to as target ...