This paper presents a case study on the IPUMS USA database that contains microdata samples sourced from censuses and surveys, encompassing a wide array of socio-economic variables such as demographics, household composition, education, employment, income, including poverty. Addressing gaps in previous studies, we propose a machine learning approach to develop several predictive models aimed at identifying and quantifying factors influencing poverty. Our experiments focus on three groups of factors: pre-disposing, socio-demographic, and socio-economic. By analysing the sensitivity of model-training variables and employing Variable Effect Characteristics (VEC) analysis to assess variable values, we evaluate the significance of various poverty-related factors.