中国农村留守儿童问题行为的因子预测：基于机器学习的方法研究

Factor Prediction of Problem Behaviors in Rural Left-Behind Children in China: A Machine Learning-Based Study

摘要: 准确预测农村留守儿童的问题行为，对于早期识别高风险个体并实施针对性干预，以改善其心理健康和社会适应具有重要意义。本研究旨在探索机器学习方法在留守儿童问题行为预测中的应用，并为相关干预措施的制定提供科学依据。本研究选取随机森林、决策树、线性回归和套索回归四种机器学习算法，对中国农村留守儿童的问题行为进行因子预测。研究首先对农村留守儿童问题行为的国内外研究现状进行综述，并深入分析了不同机器学习方法在预测问题行为方面的优势与局限性。结果表明，随机森林和决策树模型表现出较好的预测性能，经交叉验证后的平均均方误差(MSE)分别为1.056和0.012，决定系数(R²)分别为0.871和0.998。相比之下，线性回归和套索回归模型的拟合效果较差，MSE和R²值均表现不佳。综合随机森林和决策树模型的结果，识别出对问题行为预测具有重要影响的前五个因子，分别为：抑郁情绪、焦虑情绪、同伴关系、父亲教养投入以及留守儿童对未来职业的期望(期望从事工人职业)。本研究结果强调了心理健康、社会关系和家庭教养在留守儿童问题行为中的重要作用，并为利用机器学习技术进行早期风险识别和精准干预提供了新的思路和方法。

Abstract: Accurately predicting problem behaviors in rural left-behind children is of great significance for early identification of high-risk individuals and the implementation of targeted interventions to improve their mental health and social adaptation. This study aims to explore the application of machine learning methods in predicting problem behaviors among left-behind children and to provide a scientific basis for the formulation of relevant intervention measures. Four machine learning algorithms—Random Forest, Decision Tree, Linear Regression, and Lasso Regression—were selected to conduct factor prediction of problem behaviors in rural left-behind children in China. Initially, the study reviewed the current research status on problem behaviors in rural left-behind children both domestically and internationally, and conducted an in-depth analysis of the advantages and limitations of different machine learning methods in predicting problem behaviors. The results indicated that the Random Forest and Decision Tree models demonstrated superior predictive performance, with average Mean Squared Errors (MSE) after cross-validation of 1.056 and 0.012, respectively, and Coefficients of Determination (R²) of 0.871 and 0.998, respectively. In contrast, the Linear Regression and Lasso Regression models exhibited poorer fitting effects, with unsatisfactory MSE and R² values. By integrating the results of the Random Forest and Decision Tree models, the top five factors identified as having a significant impact on predicting problem behaviors were: depressive mood, anxiety mood, peer relationships, paternal parenting involvement, and left-behind children’s expectations for future occupations (expecting to work as industrial workers). The findings of this study underscore the crucial roles of mental health, social relationships, and family parenting in the problem behaviors of left-behind children and offer new insights and methods for utilizing machine learning technology for early risk identification and precise interventions.