基于多源数据和模型融合的考研成功率预测研究

Research on Predicting Postgraduate Entrance Examination Success Rate Based on Multi-Source Data and Model Fusion

摘要: 面对考研报考规模激增而录取率持续走低的现实困境，以及考生普遍存在的自我评估与志愿决策难题，本研究致力于构建精准的考研成功率预测模型。研究整合了涵盖学业、行为、心理及主观认知等多源数据，通过特征工程方法提取关键信息，并构建了基于LightGBM与CatBoost的融合预测模型。模型在真实考研学生数据上展现出优异的预测性能，F1-score和AUC指标分别达到0.78和0.88。该模型为考生提供了客观、个性化的能力评估与志愿决策支持，有助于缓解因信息不对称和集群效应导致的资源错配问题，提升研究生报考的科学性与成功率。

Abstract: Facing the dual challenges of a booming postgraduate entrance examination applicant pool alongside persistently declining admission rates, as well as the widespread difficulties candidates encounter in self-assessment and decision-making regarding institution/program choice, this paper aims to construct an accurate predictive model for postgraduate examination success probability. Our research integrates multi-source data encompassing academic performance, behavioral patterns, psychological traits, and subjective cognitive factors. Key predictive information was extracted through feature engineering methodologies, and a fusion prediction model based on LightGBM and CatBoost was developed. Evaluated on real-world data from examination candidates, the model demonstrates superior predictive performance, achieving an F1-score of 0.78 and an AUC of 0.88. This model provides candidates with objective and personalized ability assessment and decision support for institution/program selection. It contributes to mitigating resource misallocation stemming from information asymmetry and herding effects, thereby enhancing the scientific rigor and success rate of postgraduate application strategies.