Learning to Predict Rare Events in Categorical Time-Series Data

Contents

Learning to predict rare events from time-series data with non-numerical features is an important real-world problem. An example of such a problem is the task of predicting telecommunication equipment failures from network alarm data. For a variety of reasons, existing statistical and machine learning methods are not well suited to solving this class of problems. This paper describes timeweaver, a genetic algorithm based machine learning system that predicts rare events by identifying predictive temporal and sequential patterns within time-series data. Timeweaver is applied to two problems and is shown to produce results which are superior to existing learning methods.