INSHAPE: Instance-Level Shapelets for Interpretable Time-Series Classification

Time-series classification (TSC) involves categorizing data points indexed in time, which can be complex and opaque. Traditional methods focus on population-level shapelets, which are generalized patterns derived from the entire dataset, often leading to misalignment with individual time series.

INSHAPE presents a novel approach by developing instance-level shapelets that are specifically optimized for individual time series, enhancing interpretability and accuracy. The methodology involves identifying unique temporal patterns that distinguish classes at the instance level, rather than relying on broad patterns.

Results showed that using instance-level shapelets significantly improved classification accuracy compared to population-level methods, with specific metrics indicating a reduction in misclassification rates. This advancement allows for more transparent decision-making in TSC applications, making it easier for practitioners to understand model behavior.

The implications for engineers and researchers include the ability to create more tailored and interpretable models for time-series data.