An introduction to variable and feature selection
The Journal of Machine Learning Research
A review of feature selection techniques in bioinformatics
Bioinformatics
| Hi-index | 0.00 |
Themotor clinical hallmarks of Parkinson's disease (PD) are usually quantified by physicians using validated clinimetric scales such as the Unified Parkinson's Disease Rating Scale (MDS-UPDRS). However, clinical ratings are prone to subjectivity and inter-rater variability. The PD medical community is therefore looking for a simple, inexpensive, and objective rating method. As a first step towards this goal, a triaxial accelerometer-based systemwas used in a sample of 36 PD patients and 10 age-matched controls as they performed the MDS-UPDRS finger tapping (FT) task. First, rawsignalswere epoched to isolate the successive single FTmovements. Next, eighteen FT taskmovement features were extracted, depicting MDS-UPDRS features and accelerometer specific features. An ordinal logistic regressionmodel and a greedy backward algorithm were used to identify the most relevant features in the prediction of MDS-UPDRS FT scores, given by 3 specialists in movement disorders (SMDs). The Goodman-Kruskal Gamma index obtained (0.961), depicting the predictive performance of the model, is similar to those obtained between the individual scores given by the SMD(0.870 to 0.970). The automatic prediction of MDS-UPDRS scores using the proposed systemmay be valuable in clinical trials designed to evaluate and modify motor disability in PD patients.