<p>The field of time series classification has grown increasingly intricate, thereby presenting substantial challenges to achieving high classification accuracy. Although research in this domain has made significant progress, the demand for model interpretability continues to be of critical importance. Traditional Shapelet-based methods have demonstrated strong performance in terms of interpretability and classification accuracy; however, these approaches often necessitate considerable computational resources and may lead to a loss of pattern information. This study proposes an enhanced Shapelet-based model, E-STAR, which utilizes a genetic algorithm to accelerate the extraction process and incorporates supplementary frequency-domain features derived from discrete Fourier transforms. This extends traditional Shapelet extraction techniques and provides improved classification performance while preserving interpretability. Empirical evaluations on 84 UCR datasets indicate that E-STAR surpasses conventional methods. By incorporating these supplementary periodic features, E-STAR enhances interpretability without compromising accuracy, underscoring its applicability for complex time series classification tasks. For details, see the code at (https://github.com/Yxl159/E-star).</p>