Adaptive hybrid fuzzy time series forecasting technique based on particle swarm optimization

Fuzzy time series is a dynamic process in time series forecasting due to which it has gained a lot of attention from researchers. In this process, prediction accuracy is influenced by factors such as defining and partitioning the universe of discourse, fuzzification, and construction of the rule base, forecasting and defuzzification. Although numerous research have been provided in the literature, choosing the order of fuzzy time series and interval length is still a challenging task. This paper presents a computational forecasting model that overcomes the hassle of searching for the appropriate interval length and order of fuzzy time series. Particle swarm optimization is employed to search for the optimum interval length for the partitioning of the universe of discourse. Also, how changing its parameters affects the forecasting process is being investigated, which has never been done previously. A dynamic order approach is used for the selection of the order of fuzzy time series in the proposed model. In the proposed model, a sequence of orders is obtained in the training phase based upon forecast accuracy and then it is used for forecasting based upon certain rules. The model is tested on different actual time series, which include the benchmark data set of enrolments of Alabama University, the Taiwan stock exchange capitalization weighted stock index and also West Texas Intermediate crude oil prices. Different frequency datasets (e.g., yearly, monthly and daily) have been selected for this paper to check the robustness of the model. The root-mean-squared error is used as a performance parameter for the comparison of forecasting accuracy. The experimental results show that the proposed model performs better than the existing models in terms of forecasting accuracy.