Problem: Train a machine learning model to predict whether a loan will be fully paid off or “charged off” (never fully paid).
Context: Financial institutions take several variables into account when approving a loan. Determining whether a given borrower will fully pay off the loan or cause it to be charged off is difficult. If the lender is too strict, fewer loans get approved, which means there’s less interest to collect. But if they’re too lax, they end up approving loans that default. Machine learning can help us predict which loans will be charged off.
Model type: Ensemble models built with AutoML**
What we did: Using SnapLogic Data Science, we trained several models on loan data from LendingClub, a peer-to-peer lending service that has approved over 1.5 million loans since 2007. We trained the machine learning model to identify loans that are likely to end up being charged off. Banks and other lenders can use this model to avoid making bad loans and invest in good loans that yield returns. (More on how we built this demo.)
Try the Loan Repayment Prediction machine learning demo: The table below contains information on 10 approved loans from the dataset. The predictions are in the "Loan Status" column. Try changing the data and see new predictions in real-time.
Also, explore the drop-down filter in the table to the right to see how different variables (e.g., the loan amount) affect loan repayment statuses.
**The AutoML Snap (beta) automates the process of exploring different machine learning algorithms with different hyperparameters. It fully utilizes resources and delivers the best model within a specified time frame.