Desenvolvimento de modelo preditivo das repercussões da pandemia da Covid-19 na saúde mental das crianças acompanhadas no CAPSi e suas famílias com base em inteligência artificial

Abstract

The COVID-19 pandemic significantly impacted children's mental health, increasing psychosocial challenges and affecting the overall well-being of children and their families. This study aims to develop, based on Artificial Intelligence, a descriptive model (unsupervised learning) and a predictive model (supervised learning) of the repercussions of the COVID-19 pandemic on the mental health of children monitored at the Child and Adolescent Psychosocial Care Center (CAPSi) and their families. The study employed an exploratory quantitative method for generating both descriptive and predictive models. Secondary data from a previous study conducted at CAPSi in Fortaleza, Ceará, involving 150 participants aged 6 to 12 years, were utilized. The study was developed in four phases:Phase 1- Selection of variables from the secondary database; Phase 2 - Descriptive statistical analysis of the selected variables;Phase 3- Machine Learning (unsupervised learning);Phase 4 - Machine Learning (supervised learning).In Phase 1, the original database, consisting of 183 variables, was analyzed, and 48 variables were selected for evaluation in subsequent phases. Additionally, the responses of the selected variables were recategorized. In Phase 2, the selected variables were analyzed descriptively, using graphs and tables for visualization. In Phase 3, the K-means algorithm was applied in Python to create three clusters of patients. This clustering allowed for an in-depth analysis of the data and an understanding of each group. The distribution of patients across the three clusters was 29.3% for the first group (group 0), 37.3% for the second group (group 1), and 33.3% for the third group (group 2). In group 0, the quality of sleep variable emerged as a significant factor influencing the outcome, indicating a deterioration in children's mental health during the COVID-19 pandemic. In group 1, evidence suggested that some children intentionally hurt themselves, indicating worsened mental health within this group during the pandemic. In group 2, the quality of sleep variable had less impact on the outcome. When comparing groups 0 and 2, it was noted that the educational level of children in group 2 was primarily between the 4th and 6th grades, which was higher than the educational level of children in group 0. In Phase 4, predictive models based on artificial intelligence were developed using logistic regression and CatBoost. These are robust approaches for data analysis, with different capabilities for interpreting complex and categorical variables. For both models, four primary evaluation metrics were used: accuracy, precision, recall (sensitivity), and AUC (Area Under the Curve). Logistic regression achieved an accuracy of 76.6% in tests, which was considered good for the available data volume. Meanwhile, CatBoost achieved an accuracy of 80% in tests, outperforming logistic regression and demonstrating even greater robustness, particularly given the size of the dataset. This study explores the impacts of the pandemic on children's mental health using quantitative methods and artificial intelligence, providing a solid foundation for future research.