Supervised and Unsupervised Learning Approaches for Characterizing High and Low Financially Vulnerable Households

Abstract

Financial vulnerability is often described as a household's ability to cope with shocks and recover from them, and their attitude towards undertaking the risks at the household level. The recent financial and economic crisis in the pandemic stressed the deposition of financial conditions in households. We discuss in this study the analysis and identification of potential characteristics of households with higher and lower financial vulnerability scores. A survey is conducted to household heads in a local municipality in the Philippines. The data (N = 199) is first submitted to a Boruta algorithm for feature selection and out of twenty - two (22) input features, only ten (10) features are confirmed to be important in identifying lower or high vulnerability score. Afterwards, a 2 - cluster solution is revealed in the dataset using k - means clustering algorithm. Cluster 1 is formed by less educated household head with lesser household monthly debt and basic living costs, and those who are financially illiterate are more exposed to financial vulnerability. Cluster 2 is formed by household heads with higher educational attainment, high level of indebtedness and basic living costs, and those who are more financially literate and has lower financial vulnerability.