Towards Effective LLM Reasoning for Time Series Classification

The reasoning capabilities of large language models (LLMs) have significantly advanced their performance by enabling in-depth understanding of diverse tasks. However, applying LLMs to the time series domain remains nontrivial, as evidenced by the limited efficacy of directly adapting text-domain reasoning techniques. While recent work has shown promise in several time series tasks, further leveraging LLM reasoning advancements for time series classification (TSC) remains under-explored, despite its prevalence in real-world applications. In this paper, we propose ReasonTSC, a framework designed to leverages LLM reasoning capabilities for TSC through a multi-turn reasoning and a fused decision-making strategy. ReasonTSC first steers the model to think over the characteristics of time series data, integrates predictions and confidence scores from plug-in classifiers, e.g., domain-specific models, as in-context examples, and guides the LLM through a structured reasoning process: it evaluates the initial assessment, backtracks to consider alternative hypotheses, and compares their merits before arriving at a final classification. Preliminary experiments suggest that ReasonTSC can outperform both existing baselines and plug-in models, and is even capable of correcting plug-in models' false predictions.