Inductive learning of classifiers via complex fuzzy sets and logic

Mojtaba Yeganejou, Scott Dick

Abstract: Multiples studies have shown that time series forecasting algorithms based on complex fuzzy sets and logic can be both very accurate, and simultaneously very compact. There have as yet, however, been no corresponding studies of time series classification, even though it seems reasonable that similar advantages would be obtained. We propose an inductive learning architecture for time series classification based on complex fuzzy sets and logic. We evaluate this new architecture on a condition monitoring problem: detecting the onset of illness in feedlot cattle via animal-mounted sensors. We find that our new system is at least as accurate as existing approaches.