Improving the Evaluation and Actionability of Explanation Methods for Multivariate Time Series Classification
This work addresses the under-explored problem of evaluating and making explanation methods actionable for researchers and practitioners in MTSC, though it is incremental as it builds on existing methodologies.
The paper analyzed InterpretTime, an evaluation methodology for attribution methods in Multivariate Time Series Classification (MTSC), identifying weaknesses and proposing improvements to enhance accuracy and efficiency. It demonstrated the actionability of explainer rankings by applying top methods like SHAP and Feature Ablation to channel selection, achieving significant data size reduction and improved classifier accuracy.
Explanation for Multivariate Time Series Classification (MTSC) is an important topic that is under explored. There are very few quantitative evaluation methodologies and even fewer examples of actionable explanation, where the explanation methods are shown to objectively improve specific computational tasks on time series data. In this paper we focus on analyzing InterpretTime, a recent evaluation methodology for attribution methods applied to MTSC. We showcase some significant weaknesses of the original methodology and propose ideas to improve both its accuracy and efficiency. Unlike related work, we go beyond evaluation and also showcase the actionability of the produced explainer ranking, by using the best attribution methods for the task of channel selection in MTSC. We find that perturbation-based methods such as SHAP and Feature Ablation work well across a set of datasets, classifiers and tasks and outperform gradient-based methods. We apply the best ranked explainers to channel selection for MTSC and show significant data size reduction and improved classifier accuracy.