Dual Optimization of Weighted Fuzzy Time-Series Forecasting: Particle Swarm Optimization and Lagrange Quadratic Programming

Time series Forecasting is one of crucial techniques that helps with strategic decision-making and mitigating potential risks –One of which is Weighted fuzzy time series (WFTS). Moreover, the interval length of the WFTS plays a crucial role in its modelization and accuracy in predicting future values. Therefore, this research implements a dual optimization on WFTS, which are (1) Particle Swarm Optimization to find the optimum interval length of the WFTS and (2) a Lagrange quadratic to optimize the weight of the fuzzy interval. In this research, a univariate Average Air Temperature located in Malang is used to perform forecasting model. The dataset is taken from BMKG-Indonesia. This research aims to acquire an optimized interval length on fuzzy time series forecasting, i.e., improving its accuracy by finding the optimal interval length. Based on the result, the proposed dual optimization model outperforms the classical WFTS on forecasting. The proposed model excels based on the evaluation matrix values. It has been noticed also that implementing PSO to find the optimum interval length has improved the accuracy of the classical WFTS. The classical WFTS has MAPE and RMSE of 2.4 and 0.73, respectively, while the proposed dual optimized model has 1.01 and 0.3. This approach identifies the best interval values and provides optimum weights related to each data point, providing solid insights for air temperature forecasting.