Machine Learning Approaches for Predicting Individual’s Financial Inclusion Status with Imbalanced Dataset

Financial inclusion plays an important role in designing financial products and promoting sustainable development and economic growth aimed at reducing poverty. Designing financial products and services that meet individual’s financial needs requires the integration and development of advanced data-driven machine learning (ML) models. However, there is a dearth of literature on the application of intelligent ML models in predicting individual’s financial inclusion status. Therefore, this study applied random forest and decision trees to predict individual’s financial inclusion (FI) status using data collected from Hhohho, Lubombo, Manzini and Shiselweni regions in Eswatini. Among the four regions, Hhohho had the highest number of participants (27.71%), followed closely by Lubombo (25.16%), Manzini (24.20%) and 22.93% from Shiselweni region. The findings revealed that 61.15% of the participants are partially included, 33.75% are financially included, and 5.10% are financially excluded. Moreover, the synthetic minority oversampling technique (SMOTE) was used to address the class imbalance problem and further applied random forest and decision trees to predict FI status. The study results revealed that random forest performed better with a high accuracy of 98.73%, precision of 98.76%, recall of 98.73% and F1-score of 98.67%. Decision trees also performed better with an accuracy of 97.47%, precision of 98.19%, recall of 97.47% and F1-score of 97.65%. Random outperformed the decision trees across all metrics in predicting individual’s FI status. Implementing such models can aid policymakers and financial institutions in developing strategies to improve financial access and support financial inclusion initiatives.