Machine Learning for Actuaries

Decision trees, random forests, gradient boosting, neural networks, and clustering.

Key Concepts

• •Decision trees: CART algorithm, Gini impurity, entropy, and pruning
• •Random forests: bagging, feature subsampling, and out-of-bag error
• •Gradient boosting: sequential tree building with learning rate tuning
• •Neural networks: architecture, activation functions, backpropagation
• •Clustering: k-means, hierarchical clustering, and silhouette scores
• •Support vector machines: maximum margin classifiers and kernel trick
• •Model evaluation: confusion matrix, precision, recall, F1, AUC-ROC
• •Cross-validation: k-fold, leave-one-out, and stratified sampling
• •Regularization: L1 and L2 penalties for preventing overfitting
• •Feature engineering: encoding categorical variables, scaling, and interaction terms

Study Tips

• 1.Understand the bias-variance tradeoff and how each method addresses it.
• 2.Practice calculating Gini impurity and information gain for decision tree splits.
• 3.Know the hyperparameters for each method and how they affect model performance.
• 4.Compare interpretable models (GLMs, trees) with black-box models (forests, neural nets).
• 5.Focus on when to use each method rather than implementation details.

Related Exam Resources