Journal article “KFS-TUNE: Kernel-based Feature Selection for efficiency and accuracy tuning in Time Series Classification”

The EXCALIBUR Project is pleased to share the publication of the journal article “KFS-TUNE: Kernel-based Feature Selection for efficiency and accuracy tuning in Time Series Classification” in Knowledge-Based Systems.
The paper introduces KFS-TUNE, a kernel-based feature selection approach for time series classification that combines convolutional kernels with targeted feature selection. By focusing on the most informative kernel-derived features, the method aims to reduce computational cost while preserving predictive performance.

The work contributes to efficient and scalable machine-learning workflows for time series data and aligns with EXCALIBUR’s broader interest in practical, transparent, and human-centric AI pipelines.

Authors: Sofia Vei, Eleftherios Tiakas, Athena Vakali
Journal: Knowledge-Based Systems
DOI: 10.1016/j.knosys.2026.116523