Data Validation using GAIT Signals of Patients to Detect Possibility of Parkinson’s Disease

Abstract

The gait is highly individual and affected by many factors such as physique, age, gender, and emotions. Furthermore, recorded gait databases are characterized by highly dimensional, temporally dependent, highly variable, and nonlinear data vectors. Due to these reasons, automatic recognition of affect is a challenge for pattern recognition algorithms. Within this work, predictive and inferential statistics are theoretically and numerically compared. Similarities as well as dissimilarities are elaborated. Different feature extraction techniques and static as well as dynamic classification methods are compared for the recognition of emotions in gait patterns. For a small number of training samples and highly dimensional feature vectors, it is derived that the decision borders of a nearest neighbor classifier coincide with the decision borders of a support vector machine (SVM) if linear discriminant analysis is used for dimension reduction. Furthermore, algorithms trained for individuals are compared with algorithms not considering the identity of the walker. This work contributes to the state of the art by exploring various facets of pattern recognition algorithms for gait analysis and studying gait as modality for affective computing. It provides valuable insights concerning this topic and opens various perspectives for future work