High-Impact Event Prediction by Temporal Data Mining through Genetic Algorithms

This paper describes a genetic algorithm based approach to detect and predict high-impact events. While, these events occur infrequently, they are quite costly, meaning that they have a high-impact on the system key performance indicators. This approach is based on mining for these events and subsequences that are predictive of these high-impact events from historical data and then classifying these predictive patterns. The resulting mined patterns are subsequently used to make future prediction of occurrences. The approach uses a genetic algorithm for estimating the parameters for the mining process and for the prediction. This makes our approach robust as the parameters are optimized for best accuracy in classification. This approach was tested on high-impact events that occur in automotive manufacturing lines and it was found to be robust, highly accurate and with low probability of false alarms for prediction of future occurrences of such events.