Learning to predict rare events in event sequences

Abstract

Learning to predict rare events from sequences of events with categorical features is an important, real-world, problem that existing statistical and machine learning methods are not well suited to solve. This paper describes timeweaver, a genetic algorithm based machine learning system that predicts rare events by identifying predictive temporal and sequential patterns. Timeweaver is applied to the task of predicting telecommunication equipment failures from 110,000 alarm messages and is shown to outperform existing learning methods. Introduction An event sequence is a sequence of timestamped observations, each described by a fixed set of features. In this paper we focus on the problem of predicting rare events from sequences of events which contain categorical (non-numerical) features. Predicting telecommunication equipment failures from alarm messages is one important problem which has these characteristics. For AT&T, where most traffic is handled by 4ESS switches, the specific ...