Evaluation of Risk Factors for Fall in Elderly People from Imbalanced Data using the Oversampling Technique SMOTE

Abstract

Prevention of falls requires providing a small number of recommendations based on the risk factors present for a person. This article deals with the evaluation of 12 modifiable risk factors for fall, based on a selection of 45 variables from a real data set. The results of four classifiers (Logistic Regression, Random Forest, Artificial Neural Networks, and Bayesian Networks) are compared when using the initial imbalanced data set, and after using the balancing method SMOTE. We have compared the results using four different measures to evaluate their performance (balanced accuracy, area under the Receiver Operating Characteristic (ROC) curve F1-score, and F2-score). The results show that there is a significant improvement for all the classifiers when classifying each target risk factor using the data after balancing with SMOTE.