基于不完整数据的IHB-LightGBM心脏病预测模型-《中国医学物理学杂志》

文章信息/Info

Title:: Heart disease prediction using IHB-LightGBM model based on incomplete?ata

作者:: 赵小强1; 2; 3; 乔慧1; 1.兰州理工大学电气工程与信息工程学院，甘肃兰州 730050； 2.兰州理工大学甘肃省工业过程先进控制重点实验室，甘肃兰州 730050； 3.兰州理工大学国家级电气与控制工程实验教学中心，甘肃兰州 730050

Author(s):: ZHAO Xiaoqiang1; 2; 3; QIAO Hui1; 1. School of Electrical Engineering and Information Engineering, Lanzhou University of Technology, Lanzhou 730050, China 2. Key Laboratory of Gansu Advanced Control for Industrial Processes, Lanzhou University of Technology, Lanzhou 730050, China 3. National Experimental Teaching Center of Electrical and Control Engineering, Lanzhou University of Technology, Lanzhou 730050, China

Keywords:: Keywords: data mining heart disease prediction hyperparameter optimization algorithm LightGBM algorithm

摘要:: 提出基于不完整数据的IHB-LightGBM（Improved Hyperband-Light Gradient Boosting Machine）心脏病预测模型。首先，在Hyperband算法超参数采样的基础上引入了权重值，并通过蓄水池法按特征权重对其进行排序，从而筛选出最优参数以提高算法的参数寻优能力；其次，针对心脏病数据样本小且属性缺失的问题，使用K近邻算法对不完整数据进行缺失值插补，再将处理得到的完整数据进行归一化，使数据映射至0~1范围内；最后，对LightGBM采用改进后的IHB优化算法进行全局参数寻优，建立IHB-LightGBM心脏病预测模型。使用UCI心脏病数据集进行实验，结果表明IHB算法的参数寻优效果优于贝叶斯、随机搜索等优化算法，IHB-LightGBM模型在各项评价指标也上明显高于随机森林、极端随机树等算法，可以获得更快的预测速度和更高的预测精度。

Abstract:: Abstract: An improved Hyperband-light gradient boosting machine (IHB-LightGBM) model for heart disease prediction based on incomplete data is presented in the study. Based on the sampling of Hyperband algorithm for hyperparameter estimation, the weight values are introduced, and the reservoir method is used to sort the parameters according to the feature weights, so as to screen out the optimal parameters to improve the parameter optimization ability of the algorithm. Subsequently, to overcome the problems of small sample size and missing attributes of heart disease data, K-nearest neighbor algorithm is used to perform the interpolation of missing values for incomplete data, and the obtained complete data are normalized and mapped to the range from 0 to 1. The IHB optimization algorithm is adopted for global parameter optimization, and the IHB-LightGBM model for heart disease prediction is established. Experiments are carried out using the UCI heart disease data sets, and the results show that IHB algorithm is superior to Bayesian, random search and other optimization algorithms in parameter optimization, and that the various evaluation indicators of IHB-LightGBM model are significantly higher than those of random forest, extreme random tree and other algorithms. The proposed model improves both efficiency and accuracy of prediction.