Machine Learning Methods for Predicting Life Expectancy

Abstract

Forecasting of life expectancy is associated not only with serious social and financial factors, but also with the state of public health and economy, as well as with the state of the environment. The use of mathematical methods makes it possible to identify the most informative indicators affecting life expectancy. The aim of the paper is to predict life expectancy from World Bank data using machine learning (ML) methods, and to compare the effectiveness of life expectancy prediction using different machine learning algorithms, including such widely used methods as support vector method, decision tree, random forest, Fisher’s linear discriminant, neural networks, two variants of gradient bousting, logistic regression and statistically weighted syndrome method. The database included data for 238 countries. Standard non-parametric chi-square (χ²) and Mann-Whitney criteria (U-test) were applied. Eleven significant indicators were identified. Machine learning (ML) methods of Data Master Azforus data analysis system was used. The prediction result of the statistically weighted syndrome (SWS) method achieved a ROC AUC = 0.986. One-dimensional and two-dimensional diagrams of the relationship between the studied socio-economic and medical indicators on life expectancy are presented. From these charts, predictions can be derived for changes in individual indicators to improve quality and length of life. Thus, the Data Master Azforus data analysis system will enable researchers to create recommendation systems for life expectancy prediction. In addition, the conducted research will help to create a more advanced forecasting system using machine learning models that can serve as a guide for politicians makers in improving life expectancy forecasting.

Affiliated Institutions

• Institute of Biochemical Physics NM Emanuel RU
• Financial University RU

Related Publications