Abstract: Aiming at the problems of low accuracy and poor interpretability of existing heart disease risk prediction models, an interpretable heart disease risk prediction model based on improved whale optimized light gradient boosting machine (LightGBM) is proposed. First, the deep auto-encoder is used to effectively reduce the data dimensionality. Then, the whale optimization algorithm is improved by various strategies to obtain the global optimal solution of LightGBM hyper-parameter, including halton sequence initialization population, nonlinear convergence factor and dynamic spiral update. Finally, the important features of the proposed model are explained and analyzed by the shapley additive explanations method. Compared with other mainstream dimensionality reduction methods and classification models, experimental results show that the proposed model can obtain higher prediction accuracy and can efficiently extract the potential characteristics of heart disease risk factors.

Key words: heart disease risk prediction, deep auto-encoder, improved whale optimization algorithm, light gradient boosting machine, interpretability