Optimal starting point for time series forecasting

2024-09-25

arXiv.org

AI Summary

To address deteriorating forecasting performance caused by structural breaks and concept drift in time series, this paper proposes the Optimal Starting Point (OSP) strategy: it formulates starting-point selection as a data-adaptive optimization problem, enabling plug-and-play integration without modifying base model architectures. OSP jointly leverages XGBoost and LightGBM to evaluate the predictive importance of candidate starting positions within sliding windows, and employs backtesting validation to dynamically select the most robust subsequence. This work is the first to formalize starting-point selection as a temporal adaptivity decision task, thereby significantly mitigating the adverse effects of concept drift. Evaluated on multiple benchmark datasets—including M4—OSP consistently enhances state-of-the-art models such as N-BEATS and TiDE, yielding average MAPE reductions of 3.2%–7.8% over full-sequence prediction, with superior overall performance across all evaluated metrics.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Search and Optimization: Adversarial Search
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Recent advances on time series forecasting mainly focus on improving the forecasting models themselves. However, when the time series data suffer from potential structural breaks or concept drifts, the forecasting performance might be significantly reduced. In this paper, we introduce a novel approach called Optimal Starting Point Time Series Forecast (OSP-TSP) for optimal forecasting, which can be combined with existing time series forecasting models. By adjusting the sequence length via leveraging the XGBoost and LightGBM models, the proposed approach can determine the optimal starting point (OSP) of the time series and then enhance the prediction performances of the base forecasting models. To illustrate the effectiveness of the proposed approach, comprehensive empirical analysis have been conducted on the M4 dataset and other real world datasets. Empirical results indicate that predictions based on the OSP-TSP approach consistently outperform those using the complete time series dataset. Moreover, comparison results reveals that combining our approach with existing forecasting models can achieve better prediction accuracy, which also reflect the advantages of the proposed approach.

Problem

• Optimal starting point for time series
• Improving forecasting with structural breaks
• Combining OSP-TSP with existing models

Innovation

• Optimal Starting Point determination
• Integration with XGBoost and LightGBM
• Enhanced forecasting model performance