Abstract
Markov Weighted Fuzzy Time Series is a forecasting method that applies fuzzy logic to form linguistic variables from existing data. The formation of linguistic variables makes it possible for the forecasting process to be more accurate by considering the uncertainty aspect in decision-making. Its formation is started by grouping the data into a certain number of clusters. The next step is fuzzification, transition matrix formation, and defuzzification for forecasting. In the process of grouping, the existing data will be grouped into several clusters so that it results in the interval length of each cluster. One of the problems of this grouping is the absence of a base standard in the clustering process so it is prone to have a different value in forecasting accuracy. The difference in the number of the class or interval length will result in different accuracy even though the clustering method that is used is the same. In this study, the author proposes the idea of using Particle Swarm Optimization to improve the interval length. The initial interval that is already obtained through the K-means clustering algorithm will be evaluated using the Particle Swarm Optimization method so that it will have a new interval that later will be used in the fuzzification process and forecasting. The accuracy of forecasting can be calculated by using Mean Absolute Percentage Error from Markov Weighted Fuzzy Time Series conventional method and Markov Weighted Fuzzy Time Series method with Particle Swarm Optimization. The result of this study gives an improvement in error value from 8.03% to 5.88%.
