Ataxia severity classification using enhanced feature selection and ranking optimization through machine learning model

Abstract

The examination of neurological disorders and the monitoring of ataxic gait are major scientific topics that benefit from digital signal processing techniques and machine learning (ML) technologies. In this research, an ML approach is optimized with the use of Spatio-temporal data obtained from a kinect-sensor to differentiate between normal gait and ataxic. The current ML-based approaches perform very poorly because they cannot build feature-correlation among many gait characteristics. Furthermore, current ML-based techniques generate more false-positive whenever data is imbalanced in nature; especially for performing multi-label classification. This work presents a feature selection and ranking (FSR) based on extreme gradient boost (XGB) for ataxia severity classification. The FSR-XGB introduce an enhanced misclassification minimization error optimization and presents a novel feature selection and ranking to introduce feature importance using new cross-validation mechanism, both of which are aimed at solving the multi-label classification research problems. Results from experiments demonstrate that the presented FSR-XGB approach outperforms other ML-based and deep learning-based approaches.