PPGF: Probability Pattern-Guided Time Series Forecasting

• DOI: 10.48550/arxiv.2502.12802
• Publication Date: 2025-02-18
• Authors: Sun Yanru; Zongxia Xie; Hui Xing; Hualong Yu; Qinghua Hu
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Machine Learning (cs.LG)

Time series forecasting (TSF) is an essential branch of machine learning with various applications. Most methods for TSF focus on constructing different networks to extract better information and improve performance. However, practical application data contain internal mechanisms, resulting in a mixture multiple patterns. That is, the model's ability fit patterns generates errors. In order solve this problem, we propose end-to-end framework, namely probability pattern-guided time (PPGF). PPGF reformulates problem as task guided by probabilistic pattern classification. Firstly, grouping strategy approach problems classification alleviate impact imbalance Secondly, predict corresponding class interval guarantee consistency forecasting. addition, True Class Probability (TCP) introduced pay more attention difficult samples accuracy. Detailedly, classifies determine which one target value may belong estimates it accurately interval. To demonstrate effectiveness proposed conduct extensive experiments real-world datasets, achieves significant performance improvements over several baseline methods. Furthermore, TCP necessity between are proved experiments. All codes available online: https://github.com/syrGitHub/PPGF.