Fuzzy time series forecasting based on adaptive radius clustering technique

Fuzzy Time Series (FTS) techniques are gaining popularity among researchers. Fuzzification, fuzzy relation determination, and defuzzification are the three steps of fuzzy time series operations. Generally, research is focused on these stages and how to improve them. This study proposes a new clustering algorithm which partition the dataset into group by determining both the shape and the number of the clusters, and these clusters centers are used to partition the discourse. The proposed clustering algorithm is a non-parametric technique since, it uses a concept of “epsilon radius neighbours”. Also, we used the computational method to forecast the data where the weights of the forecasting parameter is optimized using the Grey-Wolf Optimization (GWO) method. Because GWO has a unique quality of striking a balance between exploitation and exploration, incorporating it into the computational method aids the model’s convergence. The model’s suitability was evaluated using data from the University of Alabama’s enrollment. The predicting accuracy of the recommended model was proved to be superior than the other models in the context of average forecasting and root mean square error. The suggested FTS forecasting method’s validity is also tested using a tracking signal (TS).