Analyzing factors linked to happiness through Machine Learning techniques

Abstract

This study aims to identify the importance of factors associated with happiness and to find patterns in the evolution of the countries that participate in the Annual world happiness report of Gallup using Machine Learning techniques. This is a longitudinal non-experimental and explanatory research of applied method (2008-2022). The work methodology used is CRISP-DM (Cross Industry Standard-Data Mining). For data imputation purposes, time series techniques were used assuming that each factor per country was a univariate time series. To identify the importance of happiness-associated factors a dependent variable level of happiness was generated from the continuous happiness variable and three other levels were proposed: high, medium, and low. During the exploratory data analysis, it was found that countries with a high level of happiness are more sensitive to the study factors. Through the decision tree model there was an average accuracy of 0.80 and three significant variables listed in order of importance: GDP per capita, life expectancy at birth and support networking. These results match with the regression by least squares with fixed effects proposed by Gallup, except in the order of the second and third factor. As for the patterns in the happiness variable evolution, the countries that participated in the last 10 years were selected, a range of techniques were applied to determine the number of clusters, and four groups were established for the clustering analysis. The results show that the first three groups maintain or improve their level of happiness while the fourth group deteriorates it.