Urban households in an emerging economy are facing financial vulnerability which is a growing socioeconomic concern and usually does not notice the traditional credit evaluation methods. This study proposes a machine learning-based model that will predict an early warning of financial vulnerability by integrating both behavioral and socioeconomic factors. A structured questionnaire was used to collect primary data comprising of 328 urban households with Indians. After data preprocess and feature engineering, three classification models were trained and compared including Logistic Regression, Random Forest, and XGBoost.The new hybrid ensemble model, based on XGBoost as a base learner and Logistic Regression as a meta-classifier, was created in order to be more predictive and interpretable. The hybrid method had the accuracy of 82 percent and ROC-AUC of 0.86. In order to be transparent of the model, SHAP (SHapley Additive explanations) analysis was conducted, and the most significant factors showed to be income, expense ratio, debt burden, and emergency savings, as well as financial confidence.At last, the model became a Flask-based web application and allowed to assess the vulnerabilities in real-time to be used practically. The research proposes a solution that is both scalable and data-driven to allow financial institutions and policymakers to actively determine the financially vulnerable households and intervene with such households before it becomes too late.
