You have labeled historical loan decisions and defaults and must build a model to predict applicant default risk. What should you do?

Training a linear regression model is appropriate because it can predict a continuous outcome (a credit default risk score) based on input features from the labeled historical data. This allows for a quantifiable risk assessment. Increasing the dataset size might be beneficial but doesn't directly address model selection for prediction. Removing bias and collecting declined applications is important for fairness and model completeness, but it's a data preparation step, not the core modeling approach for prediction. Matching social profiles for feature engineering raises significant ethical and privacy concerns, and may not be permissible or practical for a financial institution.