Identification of Determinants of Life Expectancy at Birth across Nations Using Machine Learning Techniques

Abstract

Life expectancy at birth (LEB) gives implications regarding the overall development of a nation. So identification of prominent determinant factors that affect LEB, will lead to take relevant decisions regarding the development of a nation. Studies have been conducted to identify prominent determinant factors of LEB, using ordinary least squares procedure in linear regression models with a limited number of determinant factors. Problems regarding multicollinearity, prediction accuracy and model interpretation occur when using this procedure with a wider range of determinant factors. The machine learning techniques: shrinkage and dimensionality reduction techniques were applied to overcome these problems. 17 determinant factors were identified and applied to data obtained for 193 countries of United Nations Agencies for the year 2016. As shrinkage techniques ridge and lasso regression and as dimensionality reduction techniques principal components regression and partial least squares regression were applied. These regression techniques were compared concerning mean squared error, goodness of fit and ranking based on regression coefficient estimates. Ridge regression model turned out to be the best model with a good fit for data on hand, because it has the highest adjusted R2 for the training data. Lasso regression model shows the highest adjusted R2 and lowest mean squared error for the test data. So lasso regression model is the best predictive model.