Machine Learning-Based Prediction of Nightmares in Children: A Pediatric Psychological Integration

Abstract

Identifying fear-related nightmares in children is particularly challenging, as young children often struggle to articulate their experiences or recall dream content. These nightmares, however, can severely disrupt sleep, hinder academic performance, and heighten psychological vulnerability, underscoring the need foraccurate and early prediction for effective intervention. This study explores various ML approaches, including LR, DT, SVM, KNN, NB, and LightGBM, to develop a robust predictive framework. The dataset undergoes comprehensive preprocessing, incorporating categorical encoding, numerical feature scaling, insufficient data handling, and the SMOTE to mitigate class imbalance. Each model is systematically trained and key performance indicators such as precision, recall, and F1 score were used for evaluation, with confusion matrices providing further insight into classification performance. Among the models evaluated, Light-GBM demonstrated the highest predictive performance, achieving an accuracy of 99.77%, a precision of 100%, a recall of 99.53%, and an F1 score of 99.77%, surpassing all other approaches. In this research, LightGBM is introduced as the proposed model, demonstrating superior predictive capability compared to alternative approaches and underscoring its effectiveness in predicting fear-related nightmares with high reliability. The study advances ongoing research efforts focused on machine learning applications in psychological assessment, offering a data-driven approach to early detection and facilitating timely mental health interventions.