Exploring the Explainability of Time Series Clustering: A Review of Methods and Practices

With the increasing use of time series data, particularly in critical applications and high-risk decision-making contexts, understanding and improving the explainability of time-series clustering(TSC) techniques is essential. While machine learning models excel in processing time series data, their explainability often needs enhancement, challenging human comprehension and trust. Time series data clustering, as an unsupervised learning method, extracts valuable patterns from complex datasets without prior knowledge, spanning various domains like biology and finance. However, the complexity of clustering models and their opaque decision-making processes raise concerns about understanding and trusting the results. Research in this area aims to enhance the explainability of TSC by developing new interpretation methods that not only ensure the accuracy of clustering results but also make them user-friendly and comprehensible to human users. This is crucial for overcoming challenges related to understanding and trusting the decision-making processes and outcomes of the model. In this study, we embarked on two significant endeavors: (a) We explored the use of explainable artificial intelligence (XAI) for TSC for the first time, conducting a comprehensive literature review. (b) We subdivided the research field through innovative classification methods, categorizing the explainability methods of TSC into three main categories: data preprocessing techniques based on time series data, single or hybrid methods based model training, and instance-based visualization algorithm applications. This analytical framework aims to elucidate how the explainability of TSC can be enhanced across these three dimensions, thereby improving its credibility. Our work not only opens new research avenues but also provides robust strategies for enhancing the explainability and credibility of TSC methods.