基于集成学习的风险预测模型研究与应用

Research and application of risk forecast model based on ensemble learning

针对疾病数据结构复杂以及传统模型预测精度低等问题,为探寻提高疾病预测效能的有效集成方法,提出一种集成极端梯度提升树、随机森林算法的XGB-RF预测模型,并应用于糖尿病数据集。采用网格搜索法优化模型参数,利用集成学习策略构建完整的预测模型,将多种模型的预测效果进行对比。实验结果表明,XGB-RF集成模型的准确性和解释性优于单一预测模型,可为类似疾病的早预防、早治疗提供科学、准确的辅助信息。

To explore an integrated method to improve disease prediction performance,an XGB-RF prediction model integrating logistic regression analysis,extreme gradient boosting tree and random forest algorithm was proposed,aiming at the problems of complex disease data structure and low prediction accuracy of traditional models that have been used in the prediction research of diabetic retinopathy.Grid search was used to optimize model parameters,integrated learning strategies were used to build a complete prediction model,and the prediction effects of multiple models were compared.Experimental results show that the accuracy and interpretability of the XGB-RF integrated model are better than the single prediction model,and it can provide scientific and accurate auxiliary information for the early prevention and treatment of similar diseases.