Explainable AI (XAI) for time series has seen significant algorithmic growth, yet its utility in providing measurable performance gains for downstream tasks remains under-explored. This paper introduces drXAI, a novel methodology repurposing XAI attribution methods for effective data reduction in Time Series Classification (TSC).
The core challenge in modern TSC is scalability; state-of-the-art models like Transformers exhibit quadratic complexity relative to sequence length and linear complexity relative to the number of channels, making them computationally prohibitive for massive datasets. drXAI addresses this by employing a fast, GPU-accelerated classifier (Hydra) to generate local attributions. We aggregate these into global feature importance scores and utilize an automated elbow-cut heuristic to select the most salient features without requiring manual thresholds.
Our approach is evaluated on both synthetic and real-world univariate and multivariate datasets. On synthetic benchmarks, drXAI successfully recovers ground-truth features where traditional baselines fail. On real-world data, drXAI achieves 80% to 90% data reduction while maintaining classification accuracy comparable to models trained on the full dataset. Importantly, drXAI allows resource-intensive models like ConvTran to scale to datasets previously inaccessible due to memory constraints. Our results demonstrate the benefits of using XAI not just for interpretability, but as a robust tool for feature selection and scalability in time series analysis. All our code and data are openly available.
Blogger's Review: The introduction of drXAI offers an innovative approach to data processing in time series classification, leveraging the interpretability of XAI to optimize feature selection and significantly enhance model scalability and efficiency. This represents not only an algorithmic breakthrough but also a substantial improvement in handling large datasets, showcasing the positive interaction between interpretability and performance.