Predicting Sensory Integration Disorder in 3- to 6-Year-Old Children: Application of Machine Learning Models

The aim of this study is to develop predictive models of sensory integration disorder (SID) in children, alongside identifying key factors of SID using machine learning algorithms that by establishing could help in formulating early health interventions. We conducted a stratified random sampling method to assess 1,032 3–6-year-old children (541 boys and 491 girls) using the Child Sensory Integration Scale in Fujian Province, China. Demographics, personal and family living conditions, and social environment were potential risk factors. Univariate analysis was performed to select key predictive variables. We implemented four different machine learning algorithms: eXtreme Gradient Boosting (XGBoost), Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) algorithms to build predictive models that were subsequently evaluated for performance across both training and testing data sets. The areas under the ROC curve (AUC) in the training set of 0.828, 0.758, 0.759, and 0.801 for XGBoost, LR, SVM and RF, respectively, along with an accuracy of 0.755, 0.667, 0.685, and 0.696 were achieved in the training set respectively. The AUCs in the test set was 0.725, 0.703, 0.707, and 0.690, with accuracies of 0.696, 0.686, 0.686, and 0.623, respectively. In conclusion, XGBoost was the most accurate, stable and outperformed each of the other algorithms. Sleep quality, age, screen time, emotional warmth, sleep duration were the top five factors identified by the XGBoost model. The results indicate the importance of the XGBoost model for the early detection of children with the risk of SID, which is a valid foundation for the early intervention work.