Research on the construction of an academic early warning model for vocational college students——A hybrid model based on Logistic regression and random forest
高职院校学生学业预警模型构建研究——基于Logistic回归与随机森林组合模型

摘要

当前国内学业预警研究多聚焦本科院校，针对高职院校的量化建模较为匮乏，普遍存在预警滞后、指标体系缺乏职业教育特色、模型可解释性与预测精度难以兼顾等问题。本文以高职院校学生为研究对象，结合高职院校工学结合、以实训实习为主的办学特点，融入实训成绩、岗位实习表现等特色指标，构建基于Logistic回归与随机森林的组合学业预警模型。首先运用Logistic回归结合LASSO算法完成特征筛选，明确学业风险核心影响因素；再以随机森林作为主预测模型提升预警精度，并通过XGBoost、单一决策树开展对比验证。依托AUC值、准确率、召回率等指标完成模型评估，最终划分红、黄、蓝三级学业风险预警等级，配套制定分层分类干预策略，形成从预警到干预再到反馈的闭环管理机制。

Abstract

Most existing domestic studies on academic early warning focus on undergraduate universities, while quantitative modeling for vocational colleges is insufficient. Common problems include delayed early warning, an indicator system lacking features of vocational education, and difficulties in balancing model interpretability and prediction accuracy. Taking vocational college students as the research subjects, this paper conforms to the school-running characteristics of vocational colleges featuring work-integrated learning and prioritizing practical training and internships, incorporates characteristic indicators such as practical training scores and post internship performance, and constructs a hybrid academic early warning model based on Logistic Regression and Random Forest. Firstly, Logistic Regression combined with the LASSO algorithm is adopted for feature selection to identify the core influencing factors of academic risks. Random Forest is then used as the main prediction model to improve early warning accuracy, with XGBoost and a single decision tree applied for comparative verification. The model is evaluated by indicators including the AUC value, accuracy rate and recall rate. Finally, three tiers of academic risk warning levels (Red, Yellow and Blue) are divided, with hierarchical and classified intervention strategies formulated correspondingly, so as to form a closed-loop management mechanism covering early warning, intervention and feedback.