A hybrid random forest model for stroke risk prediction

用于预测中风风险的混合随机森林模型

阅读:1

Abstract

OBJECTIVES: Our study aims to develop a stroke risk prediction model by multiple machine learning algorithms and optimize the model as a stroke risk prediction tool. METHODS: This retrospective multicenter study derived the original dataset from a high-quality health database. The dataset was incomplete and class imbalanced. Firstly, we eliminated extreme outliers and noises and imputed missing values by appropriate algorithms. We further used Synthetic Minority Over-sampling Technique to generate a balanced dataset. Secondly, we fitted seven algorithms to develop a machine learning-based prediction tool for clinical practice. RESULTS: Overall, 35,859 participants were included, of whom 781 (2.2%) experienced a stroke. The random forest model demonstrated the best performance with high predictive value and discrimination ability. For stroke risk prediction, the AUC of the best-performing model was 0.97. CONCLUSION: A new random forest algorithms-based stroke risk prediction model using easily obtainable data was developed and outperformed established models. Future studies should further validate and optimize the current model to assess its generalizability and promote the wide application. The utilization of proposed random forest algorithms as an individualized risk prediction model could facilitate the application of clinical practice guidelines and shared decision-making.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。