Identification of self-care problem in children using machine learning

Abstract

Identification of self-care problems in children is a challenging task for medical professionals owing to its complexity and time consumption. Furthermore, the shortage of occupational therapists worldwide makes the task more challenging. Machine learning methods have come to the aid of reducing the complexity associated with problems in diverse fields. This paper employs machine learning based models to identify whether a child suffers from self-care problems using SCADI dataset. The dataset exhibited high dimensionality and imbalance. Initially, the dataset was converted into lower dimensionality. Imbalanced dataset is likely to affect the performance of machine learning models. To address this issue, SMOTE oversampling method was used to reduce the wide variations in the class distribution. The classification methods used were Naïve bayes, J48 and random forest. Random forest classifier which was operated on SMOTE balanced data obtained the best classification performance with balanced accuracy of 99%. The classification model outperformed the existing expert systems.

Keywords

• Self-care problem
• machine learning
• Imbalanced data
• Classification