Understand your loan decision.
Enter applicant details to estimate approval probability and understand which financial factors influenced the model's prediction through mathematical feature contributions.
How CreditWise Explains Decisions
A look into the underlying feature engineering and explainability mathematics.
1. Preprocessing & Imputation
Missing numerical fields are imputed using dataset means, while missing categorical values are imputed using mode frequency. Categorical variables are encoded with LabelEncoder and OneHotEncoder with first-category dropping to prevent collinearity.
2. Non-linear Feature Engineering
Financial risk exhibits non-linear curvature: risk escalates rapidly at high debt ratios and drops significantly at high credit scores. The pipeline introduces DTI_Ratio2 and Credit_Score2 to model these dynamics inside a linear model.
3. Feature Contributions
Features are standardized with StandardScaler. The log-odds contribution of each feature is mathematically defined as: cj = βj × xscaledj. Positive values push the probability toward approval, while negative values push toward rejection.
4. Conceptual Aggregation
Rather than presenting confusing dummy variables or squared terms, CreditWise aggregates one-hot dummies and maps technical features to plain-language financial drivers.