Fuzzy time series forecast with enhanced trends and weighted defuzzification

Abstract

Fuzzy time series is a constructive device to forecast imprecise and vague data in real-life scenarios. In this process, main steps involved are fuzzification, construction of fuzzy logic relationships and defuzzification. The results of forecasting can be enhanced by amendments in its steps. In this paper, author presents a method for fuzzy time series forecasting based on enhanced method for construction of rule base and modified weighted defuzzification. A grouped rule base is formed using fuzzy logical relationship of high order (first, second and third) and order based weighted approach is proposed for defuzzification process. To test the performance of the presented method, it is applied to a benchmark problem of student's enrolments of Alabama University. The accuracy is analysed with the help of comparative study of proposed method with other existing techniques. Performance evolution criteria indicate the applicability of proposed method.