Predictive Model Plan – Delinquency
Prediction
1. Model Logic (Generated with GenAI)
Objective:
The model predicts whether a customer is likely to become delinquent (0 = No, 1 = Yes).
Model Workflow:
• Handle missing values using imputation.
• Encode categorical variables such as Employment_Status, Credit_Card_Type, and
Location.
• Scale numerical variables if necessary.
• Create additional features from payment history (Month_1 to Month_6).
• Split the dataset into 80% training and 20% testing data.
• Train a Logistic Regression model.
• Predict delinquency probabilities and classify customers into risk categories.
Pseudo-code:
Load dataset
Clean missing values
Encode categorical variables
Split data into train and test sets
Train Logistic Regression model
Predict delinquency probability
Evaluate model performance
Brief Explanation:
The model uses customer financial behavior, credit history, employment status, and
payment records to predict delinquency risk.
2. Justification for Model Choice
Selected Model: Logistic Regression
Reasons:
• Suitable for binary classification because Delinquent_Account has two outcomes.
• Highly interpretable and transparent for financial institutions.
• Easy to implement and computationally efficient.
• Generates probability scores useful for risk segmentation.
• Supports explainable AI and regulatory compliance.
Suitability for Geldium:
The model enables reliable and explainable credit risk assessment, helping Geldium make
informed lending decisions.
3. Evaluation Strategy
Performance Metrics:
• Accuracy: Overall correctness of predictions.
• Precision: Percentage of predicted delinquent customers that are truly delinquent.
• Recall: Percentage of actual delinquent customers correctly identified.
• F1 Score: Balance between precision and recall.
• AUC-ROC: Measures the model's ability to separate classes.
Interpretation:
• High recall reduces the risk of missing delinquent customers.
• Precision minimizes false alarms.
• Higher AUC-ROC indicates better model performance.
Bias Detection and Reduction:
• Evaluate performance across demographic groups.
• Monitor fairness metrics such as demographic parity and disparate impact.
• Address class imbalance using resampling methods if required.
Ethical Considerations:
• Protect customer privacy and data security.
• Avoid discriminatory outcomes.
• Use predictions as decision-support tools rather than fully automated decisions.
• Ensure transparency and explainability.
4. Final Submission
A Logistic Regression model is proposed to predict customer delinquency risk using
financial, behavioral, and credit-related variables. The model is interpretable, efficient,
and suitable for binary classification tasks. Model performance will be evaluated using
Accuracy, Precision, Recall, F1 Score, and AUC-ROC while ensuring fairness and
ethical AI practices.