Prediction Of Problem Behaviors Of Migrant Children Based On Machine Learning

Migrant children are a special group in Chinese society that requires close attention.In recent years,scholars have explored the influencing factors of their problem behaviors,and the factors that affect migrant children's problem behaviors were complex and diverse.Common predictable behaviors of migrant children included interpersonal relationships,family socioeconomic status,and family relationships.This research intended to explore the influencing factors of migrant children's problem behaviors,and established a predictive model through machine learning algorithms to discover important factors that affect migrant children's problem behaviors.In the literature review section,the related theories of problem behaviors and related researches were reviewed,including a series of demographic variables,positive/negative emotions,social anxiety,perceptions of discrimination,loneliness,subjective family support,family socioeconomics status,teacher-student relationship,social resources and life events of adolescents,and sorted out relevant research on the application of machine learning.In this study,3126 migrant children from the third to fifth grades of 11 schools in Beijing were selected as subjects,and SPSS23.0,R 3.6.2 and Python 3.7 were used to analyze and model the research data.Research results:(1)The scores of migrant children's problem behaviors,perceptions of social discrimination,life stress events,teacher-student relationship attachment and intimacy,social anxiety and fear of negative evaluation,and loneliness are all at the upper middle level.The scores of family discrimination perception,positive emotion,negative emotion,school discrimination perception,social resources,social anxiety,social avoidance and distress,conflict and avoidance of teacher-student relationship,family socioeconomic status and problem behavior are all in the lower middle level.(2)There are significant differences in the problem behaviors of migrant children in gender,grade,and family structure.(3)Using machine learning models to discovered that important factors affecting migrant children's problem behaviors include life stress events,negative emotions,teacher-student relationship,subjective family support,loneliness,family socioeconomic status and positive emotions.(4)When the machine learning model predicts problem behaviors of migrant children,its prediction performance could reach an average of 86.7% AUC value,77.88%precision,79.5% accuracy,77.57% recall and 78.51% F1 and a Brier score of 0.146.Research conclusions:(1)The influencing factors of migrant children's problem behaviors were multifaceted and multifactorial.(2)The use of machine learning to established a predictive model of migrant children's problem behavior was feasible and had strong predictive ability.This study proposed educational suggestions from the individual,family,school,and social levels to reduce migrant children's negative emotions,social anxiety,perceptions of discrimination,loneliness and teenage life events,and increase migrant children's subjective family support,family socioeconomic status,teachers and students relationships,social resources,and positive emotions,thereby reducing problem behaviors.