Automated Loan Eligibility Prediction using H2O.ai
Project Overview
Built a machine learning model using H2O.ai to automate loan eligibility screening, helping a Kenyan lender predict applicant approval likelihood from application data in real time rather than relying on fully manual review.
The Challenge
Loan approval is core to a lender's business, but manual verification of every applicant against income, credit history, employment, and demographic factors is slow and inconsistent between reviewers. The lender needed a way to automate a first-pass eligibility screen so staff could focus review time on borderline or high-risk applications rather than routine ones.
Our Solution
Trained a classification model using H2O.ai's AutoML capability on applicant data (income, marital status, education, credit history, dependents, loan amount, property area) to predict loan approval likelihood. H2O AutoML was used to automatically compare multiple algorithms (GLM, GBM, Random Forest, Stacked Ensembles) and select the best-performing model without manually tuning each one, speeding up the build-to-deployment timeline.
Impact & Results
Enabled real-time, model-assisted loan eligibility screening, turning applicant data into actionable approval-likelihood estimates to support initial assessment. By applying a consistent screening process across applications, the solution helps lending staff prioritise their workload and direct closer attention to borderline or complex cases requiring human judgement. The automated approach provides a foundation for reducing repetitive screening work, shortening initial assessment turnaround times, and handling growing application volumes more efficiently. It also creates opportunities to monitor prediction quality against actual lending decisions and refine the model as new data becomes available. Together, these capabilities support a more scalable, data-informed lending workflow, with eligibility predictions serving as decision support for the lender’s review process.