In this paper we propose a method to extract frequent temporal patterns from medical time series. We extend a conventional interval-based mining approach to embed fuzziness on the duration of a temporal event and the gap between temporal events. We also allow a case to support multiple temporal relations at respective membership degrees in order to alleviate the decrease of support in finding frequent patterns and make it possible to incorporate ranges in temporal relations. Consequently, our

more »

... thod generates frequent patterns with abstracted ranges such as 'A co-occurs several days with B' and 'A occurs several weeks after B', which can be used to represent the temporal evolution of a disease.