Angular correlation-based feature selection for machine learning classification of manual automatisms using body sensor network data.

Automatisms are repetitive, semi-ordered movements often observed in focal impaired awareness seizures and, less frequently, in generalized seizures with brief loss of consciousness. This study aims to improve the detection of these automatisms by optimizing the feature selection step in a machine learning system. An Angular Correlation Algorithm (ACA) is proposed to select relevant statistical features from inertial data gathered by a five-module body sensor network. The ACA enhances efficiency by identifying significant features that support high classifier accuracy. When compared to traditional methods like one-way ANOVA, ACA effectively selects 80 % of the relevant features, outperforming ANOVA's 67.85 %, and requires less processing time and power. The findings suggest that ACA not only improves detection accuracy but also streamlines the feature selection process for seizure-related automatisms. The study concludes that ACA offers a valuable, efficient approach while maintaining high accuracy, demonstrating the effectiveness of the proposed method.