Ensemble Modeling Method to Predict Life Expectancy of Population in High-Income Countries: Japan and Finland

Life expectancy at birth is an indicator defined by the United Nations Development Program (UNDP) as the number of years, on average, an infant is expected to live. This indicator is a proxy of good health. The health index together with the education and income indices are used by UNDP for measuring the development level of the member countries. In addition to improve human development along the health dimension, most governments also need the accurate projection of life expectancy of their populations for the effective social services and decent pension planning. In this work, we propose a data-driven modeling method to predict life expectancy. Our method is based on the ensemble scheme in which a combination of classification and regression tree (CART) and the chi-square automatic interaction detection (CHAID) algorithms are applied for making a cooperative prediction. We empirically prove that the proposed ensemble scheme is more accurate than a single model prediction. We experiment our modeling methodology with the life expectancy data of the two high-income countries: Japan and Finland. This selection is due to the fact that these two countries are in the group of very high human development according to the latest UNDP ranking report. The CART and CHAID models reveal that both economic and environmental factors share their contributions to forecasting life expectancy of populations in the two countries. Forest depletion, agricultural methane and CO 2 emissions, particulate emission damage, national income, and education expenditure are factors affecting longevity of Japanese population. To predict the Finn's life expectancy, the ensembled models consider several factors including exports and imports of goods and services, electric power consumption, energy use, national income, GDP growth, education expenditure, forest area, agricultural methane emission, and particulate emission damage.