Comparison of Predicting Regional Mortalities Using Machine Learning Models

Prediction of mortality is an important problem for making plans related to health and insurance systems. In this work, mortality of Africa, America, East Asia and Pacific, Europe and Central Asia, Europe alone, South Asia regions have been studied and predictions are made using fourteen machine learning techniques. These are linear, polynomial, ridge, Bayesian ridge, lasso, elastic net, k-nearest neighbors, support vector (with linear, polynomial and radial basis function kernels), decision tree, random forest, gradient boosting and artificial neural network regressors. The results are compared based on the coefficient of determination and the accuracy values. The best predicting algorithm varies from one region to another. On the other hand, the best accuracy (99.32%) and coefficient of determination (0.9931) are obtained for Africa region and using k-nearest neighbor regressor.