Research Specifications

Recently, clustering has been widely used to localize global forecasting models and address data heterogeneity. Most prior work has focused on single-view clustering approaches, where constructing effective representations of time series data is critical. Inspired by advances in multi-view clustering and time series imaging techniques, and aiming to leverage the complementary information captured from multiple views of time series data, this study proposes a global forecasting methodology based on multi-view clustering (MVC) to enhance clustering robustness and improve forecasting accuracy. Specifically, we introduce an optimal fuzzy multi-view clustering algorithm (OFMVC) as part of the proposed forecasting framework. OFMVC includes an intelligent view selection mechanism that leverages diversity and manifold learning to identify views that are both complementary and representative of the dataset’s inherent geometric structure. Experiments on the six M3 monthly datasets show that forecasting models built using OFMVC outperform all single-view clustering methods. In terms of mean symmetric Mean Absolute Percentage Error (sMAPE), the proposed model also exceeds a baseline multi-view clustering method, with a performance gain of 0.21. Moreover, the proposed approach outperforms several state-of-the-art methods reported in the literature, including univariate and deep learning-based time series forecasting models. The source code and data for this article are located at: https://github.com/AliReza000J/MVC-Time-series-Forecasting.