Enhanced Financial Inclusion with Boosting-Tree Transfer Learning

Around the world, many people still don’t have access to financial services, and this issue is especially common in countries with lower or moderate incomes where it is difficult for marginalised groups to obtain official financial services. In order to improve predictive modelling for financial inclusion, this work explores the use of cutting-edge machine learning techniques, including boosting-tree transfer learning. The study uses mobile money transaction records and publicly accessible financial survey data, combining behavioural, demographic, and socioeconomic characteristics. Transfer learning is used to adjust models across regions with restricted data availability, whereas gradient boosting decision trees (GBDT) are utilised as the base learners because of their excellent capacity to model nonlinear relationships. The model was tested using stratified kfold cross-validation, and its performance was measured with metrics like area under the ROC curve (AUC), accuracy, precision, recall, and F 1 -score. The outcomes show that boosting-tree transfer learning achieves 90% accuracy and an AUC of 0.92 on test sets, outperforming baseline models like logistic regression and random forests. Crucially, there are few differences across age and gender groups in fairness-aware judgement, which lowers the possibility of algorithmic exclusion. According to the results, boosting-tree transfer learning offers a scalable strategy to promote inclusive financial systems around the world by enhancing predictive accuracy and facilitating equitable decision-making.

Ҳали таржима қилинмаган