Optimizing Shapelet Lengths for Effective Time Series Classification

In this paper, we propose a novel framework for optimizing shapelet lengths to enhance time series classification accuracy. Our approach involves a systematic exploration of shapelet lengths, combined with a multi-objective optimization strategy that balances model complexity and classification accuracy. We introduce a method for dynamically selecting shapelet lengths based on data characteristics, allowing for a more tailored extraction process. We evaluate our framework on several benchmark time series datasets, comparing its performance against traditional shapelet-based methods and other classification techniques. Our results demonstrate that optimizing shapelet lengths leads to significant improvements in classification accuracy and robustness, particularly in datasets with varying temporal patterns. This work contributes to the field of time series analysis by addressing a critical gap in shapelet-based methodologies. By providing a structured approach to shapelet length optimization, we aim to enhance the applicability of shapelet methods across diverse domains, paving the way for more accurate and interpretable time series classification solutions.