Feature selection for relative poverty classification in Morocco using supervised machine learning algorithms

This study examines the determinants of relative poverty in Morocco using a supervised machine learning approach. Using nationally representative data from the 2013–2014 National Household Consumption and Expenditure Survey, a random forest classifier is employed to identify the most influential factors distinguishing poor from non-poor households. The results highlight the multidimensional nature of relative poverty, with education, household composition, housing costs, and regional disparities emerging as key predictors. In particular, illiteracy, household size, and rental expenditures demonstrate strong explanatory power. These findings provide data-driven insights to improve poverty targeting and inform multidimensional social policies in Morocco, while also offering a transferable framework for poverty diagnosis in other countries.