CLASSIFICATION OF PATIENTS WITH VISUAL DISABILITY FOR THE BASIC FUNCTIONAL REHABILITATION PROGRAM

Abstract

The rehabilitation needs of people with visual disabilities vary due to different factors: congenital or acquired eye diseases, new social interaction, adaptation to productive, academic, or similar life, among others, meaning a continuous readjustment of the treatment plan of the patient and increasing rehabilitation costs. The care history of the Rehabilitation Center for Blind Adults (CRAC) has followed the basic functional rehabilitation route of the Rehabilitation Manual of the Latin American Union of the Blind (FOAL-ULAC, 1999) and contains the variables that can efficiently classify these patients with the use of machine learning tools.The patient's demographic (gender, age group) and clinical data (visual condition, admission ophthalmological diagnosis, emotional state), along with the number of visits, were collected through non-probabilistic sampling; The multiple correspondence analysis resulted in inverse associations between the quantitative and categorical variables, there was also a positive correlation between the quantitative variables according to Pearson's coefficient. To define the classification target variable, the similarcharacteristics between the variables were grouped into two clusters, using the scikit-learn library in Python and the k-prototype algorithm; After having the objective-labeled variable, the supervised Decision Tree, Random Forest, Gradient Boosting and Logistic Regression models were trained and tested. These models gave an accuracy between 82% and 84%, the most effective being the Gradient Boosting model, whose class prediction was: of 516 True Positives, 11 are False Positives, of 94 True Negatives and 15 are False Negatives (recall 98%), and accuracy of 83%.Keywords: Basic Function Rehabilitation, Supervised Machine Learning, Clustering, Predictive Models