近年来，“校园欺凌”逐渐出现在大众的视野中，校园欺凌现象屡见不鲜。由校园欺凌引发的问题逐渐增加，使得校园欺凌影响因素分析成为教育界关注的问题。因此，研究初中生校园欺凌行为的影响因素，可以有效地对校园欺凌行为进行防控，并有效地减少由校园欺凌行为引发的一系列对学生的伤害现象。

本文基于中国教育追踪调查2014-2015年的数据，构建能够预测初中生校园欺凌行为的机器学习模型，并对模型效果进行评估，选出效果最优的机器学习模型，并根据生成的特征重要性排序图来探究影响初中生校园欺凌行为的因素。

本研究发现，在逻辑回归、决策树、随机森林和神经网络这四种模型中，随机森林模型的效果最好，精度达到了90.2%。研究发现较为重要的影响因素有学生的自我调控能力、学校归属感、心理健康状况以及教师关心等，学生是否早恋也是值得关注的一个因素。在不同性别的学生中，男生的欺凌行为更易受到其自我调控能力和自身心理健康状况的影响，而女生的欺凌行为则更受各种因素的综合作用。对各因素的影响方向进行探究，发现除同伴不良行为的影响为正向外，心理健康状况、同伴积极行为、教师关心、父母行为参与时间和学校归属感等对学生实施欺凌的影响均为负向，即同伴不良行为水平越高的学生更易实施校园欺凌行为，而心理健康状况越好、同伴积极行为水平越高、受到的教师关心越多、父母行为参与时间越多以及学校归属感越强的学生越不容易实施校园欺凌行为。本研究中的建模方法更关注相对较积极的变量，这提醒我们，想要降低实施校园欺凌的风险，可以先从给学生营造积极的环境入手。

In recent years, "school bullying" has gradually emerged in the public eye, and the phenomenon of school bullying is not uncommon. The problems caused by school bullying are gradually increasing, making the analysis of the influencing factors of school bullying a concern in the education industry. Therefore, studying the influencing factors of school bullying behavior among junior high school students can effectively prevent and control school bullying behavior, and effectively reduce a series of harm phenomena caused by school bullying behavior to students.

This article is based on the data from the China Education Panel Survey from 2014 to 2015, and constructs machine learning models that can predict whether middle school students engage in school bullying behavior. After evaluating the model's effectiveness, the machine learning model with the best performance is selected. Based on the generated feature importance ranking chart, the factors that affect middle school students' school bullying behavior are explored.

This study found that among the four models of logical regression, decision tree, random forest and neural network, random forest model had the best effect, with an accuracy of 90.2%. Research has found that the important influencing factors include students' self-regulation ability, sense of belonging to school, mental health status, and teacher care. Whether students have early love is also a factor worth paying attention to. Among students of different genders, male bullying behavior is more susceptible to the influence of their self-regulation ability and mental health status, while female bullying behavior is more influenced by a combination of various factors. Exploring the impact direction of various factors, it was found that the impact of peer negative behavior is positive, and the impact of mental health status, peer positive behavior, teacher care, parental behavior participation time, and school sense of belonging on students' bullying behavior was all negative. That is, students with higher levels of peer negative behavior were more likely to engage in school bullying behavior, while students with better mental health status and higher levels of peer positive behavior were found to have a negative impact. Students who receive more attention from teachers, spend more time participating in parental behavior, and have a stronger sense of belonging to the school are less likely to engage in school bullying behavior. The modeling method in this study focuses more on relatively positive variables, which reminds us that to reduce the risk of implementing school bullying, we can start by creating a positive environment for students.