Improving the Evaluation and Actionability of Explanation Methods for Multivariate Time Series Classification

TL;DR

This paper critically evaluates and improves methods for explaining multivariate time series classifiers, demonstrating that better explanations can enhance feature selection and classifier performance.

Contribution

It identifies weaknesses in existing evaluation methodologies and proposes improvements, while also demonstrating how explanation methods can be used for actionable channel selection in MTSC.

Findings

Perturbation-based attribution methods outperform gradient-based ones.

Using explanations for channel selection reduces data size and improves accuracy.

Improved evaluation methods increase the reliability of explanation techniques.

Abstract

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.