Learning Objectives:

  • Explain unsupervised learning applications in banking.

  • Understand customer segmentation and anomaly detection.

  • Explain ensemble methods and their advantages.

4.1 Unsupervised Learning in Banking

Unsupervised learning finds patterns in unlabelled data. Key applications include:

  • Customer Segmentation: Grouping customers by behaviour or needs.

  • Anomaly Detection: Identifying unusual transactions (fraud).

  • Dimensionality Reduction: Simplifying complex data.

The HKU SPACE course introduces fundamental ML concepts, including the basics of machine learning, neural networks, and deep learning architectures . The NUS course covers Supervised and Unsupervised Learning techniques .

4.2 Ensemble Methods

Ensemble methods combine multiple models to improve performance:

  • Bagging: Reduces variance by training models on different subsets (e.g., Random Forest).

  • Boosting: Reduces bias by training models sequentially, focusing on misclassified examples (e.g., XGBoost, Gradient Boosting).

  • Stacking: Combines predictions from multiple models.

The ETH Zürich course covers bagging, random forests, and gradient boosted trees (XGBoost) . XGBoost is particularly popular in credit scoring competitions due to its high accuracy.

4.3 Dimensionality Reduction

Dimensionality reduction techniques reduce the number of features while preserving key information:

  • Principal Component Analysis (PCA): Linear dimension reduction.

  • Singular Value Decomposition (SVD): Decomposing a matrix into its component parts.

  • Autoencoders: Neural network-based dimension reduction.

The ETH Zürich course covers dimension reduction methods, singular value decomposition, and autoencoders .

 

PreviousNext