Multiple Linear Regression in Predicting Motor Assessment Scale of Stroke Patients

Abstract

The Multiple Linear Regression (MLR) is a predictive model that was commonly used to predict the clinical score of stroke patients. However, the performance of the predictive model slightly depends on the method of feature selection on the data as input predictor to the model. Therefore, appropriate feature selection method needs to be investigated in order to give an optimum performance of the prediction. This paper aims (i) to develop predictive model for Motor Assessment Scale (MAS) prediction of stroke patients, (ii) to establish relationship between kinematic variables and MAS score using a predictive model, (iii) to evaluate the prediction performance of a predictive model based on root mean squared error (RMSE) and coefficient of determination R2. Three types of feature selection methods involve in this study which are the combination of all kinematic variables, the combination of the best four or less kinematic variables, and the combination of kinematic variables based on p < 0.05. The prediction performance of MLR model between two assessment devices (iRest and ReHAD) has been compared. As the result, MLR model for ReHAD with the combination of kinematic variables that has p < 0.05 as input predictor has the best performance with Draw I (RMSEte = 1.9228, R2 = 0.8623), Draw Diamond (RMSEte = 2.6136, R2 = 0.7477), and Draw Circle (RMSEte = 2.1756, R2 = 0.8268). These finding suggest that the relationship between kinematic variables and MAS score of stoke patients is strong, and the MLR model with feature selection of kinematic variables that has p < 0.05 is able to predict the MAS score of stroke patients using the kinematic variables extracted from the assessment device.

Linear and Non-Linear Predictive Models in Predicting Motor Assessment Scale of Stroke Patients Using Non-Motorized Rehabilitation Device

“…Multiple Linear Regression (MLR) is the most fundamental and straightforward approach for experimental analysis and data processing in analytical contexts. It serves as a sophisticated statistical approach for elucidating correlations between multiple input predictors and a single response variable [36,37]. The response variables y exhibit linear correlations with multiple predictor variables.…”