Matrix Profile-Based Interpretable Time Series Classifier.

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Time series classification (TSC) is crucial across domains like healthcare and finance.
• Current AI models for TSC lack interpretability, limiting their use in sensitive applications.
• Existing explanation methods for black-box models are a workaround, not an ideal solution.

Purpose of the Study:

• To develop a novel time series classification method that is transparent by design.
• To create an efficient and effective interpretable TSC approach.

Main Methods:

• Proposes an interpretable TSC method leveraging patterns extracted from the Matrix Profile (MP) of training time series.
• Designs a classification procedure modeled as a decision tree for transparency.
• Identifies discriminative subsequences as the basis for classification decisions.

Main Results:

• The proposed method demonstrates superior performance compared to existing state-of-the-art interpretable TSC approaches.
• Quantitative and qualitative experiments validate the effectiveness and efficiency of the transparent classifier.
• The decision tree explicitly shows classification reasoning through subsequence presence.

Conclusions:

• The novel Matrix Profile-based method provides an interpretable, efficient, and effective solution for time series classification.
• This approach addresses the critical need for transparency in AI models for sensitive domains.
• The method outperforms current interpretable alternatives, offering a significant advancement in the field.