关键词: 特征选择, bp-SVM, 死亡率预测, 心衰, 机器学习

Abstract: This paper constructs a mortality prediction system that predicts mortality within 30 day after hospitalization in heart failure patients. Data are collected from the Shanghai Shuguang hospital. It preprocesses the data and feature of heart failure patients. A traditional feature selection algorithm called Relief is utilized to select important feature because of redundancy of feature. Finally, this system adopts biased penalties-SVM（bp-SVM） algorithm to predict mortality for heart failure patients. The experimental results show that system can achieve high performance in 30-day mortality prediction. Mortality prediction system aims to assist doctors in treating heart failure patients by providing important decision information. Doctors can apply individualized treatment according to different condition of patients in order to improve the result of clinical diagnosis and the resource allocation.

Key words: feature selection, biased penalties-Support Vector Machine（bp-SVM）, mortality prediction, heart failure, machine learning