Learning Objectives:

  • Explain supervised learning algorithms and their applications in banking.

  • Understand regression models for continuous outcomes.

  • Understand classification models for categorical outcomes.

3.1 Regression Models in Finance

Regression models predict continuous outcomes. They are widely used in finance for:

  • Stock Return Prediction: Predicting future stock prices.

  • Insurance Premium Calculation: Estimating insurance premiums.

  • Risk Modelling: Quantifying financial risk.

The ETH Zürich course includes linear, polynomial, ridge, and lasso regression as core topics . Linear regression is often used as a baseline model, while regularisation techniques (ridge and lasso) help prevent overfitting . The RPTPU course covers linear models with L2 loss (ridge) and L1 loss (lasso), along with practical sessions on implementing these models in Python .

3.2 Classification Models in Finance

Classification models predict categorical outcomes. They are used for:

  • Credit Scoring: Classifying borrowers as “default” or “non-default”.

  • Fraud Detection: Classifying transactions as “fraudulent” or “legitimate”.

  • Customer Churn Prediction: Identifying customers likely to leave.

The ETH Zürich course includes logistic regression as a core algorithm . The University of Warsaw course covers logistic regression, K-nearest neighbours (KNN), Support Vector Machines (SVM), decision trees, and Random Forest . Random Forest is an ensemble method that uses bagging (bootstrap aggregating) to improve predictive accuracy .

3.3 Model Evaluation

Evaluating model performance is critical. Key evaluation metrics include:

  • Regression Metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R-squared.

  • Classification Metrics: Accuracy, Precision, Recall, F1-Score, AUC-ROC.

The RPTPU course covers performance evaluation metrics for both regression and classification problems . The University of Warsaw course includes assessing model accuracy, learning curves, cross-validation, and the concept of bias-variance trade-off .

 

PreviousNext