STAT-5525: Statistical Learning

Description: Theory and application of supervised and unsupervised methods of statistical and machine learning. 5525: Methods of supervised statistical and machine learning for regression and classification. Overview of statistical (data) and algorithmic models. Detailed study of regression models for continuous and discrete data (linear, nonlinear, and generalized linear models). Detailed study of methods for classifying categorical outcomes (logistic and multinomial models, discriminant analysis, naïve Bayes). Tree-based methods for regression and classification. Feature selection, regularization, and dimension reduction for high-dimensional problems (Lasso, Ridge, PCR, PLS). Cross-validation and resampling for model tuning and uncertainty estimation. Statistical analyses using R or Python. 5526: Supervised and unsupervised statistical and machine learning for complex or high-dimensional data. Methods include: global and local models with smoothing (nearest-neighbor, kernel, and basis expansion techniques). Generalized (linear and additive) models. Methods for correlated (clustered) data, mixed models. Unsupervised learning for summarization, visualization, dimension reduction, imputation, and grouping; (K-means, PCA, hierarchical clustering, model-based clustering, association rules, self-organizing maps, and biclustering). Ensemble learning (bagging, model averaging, boosting, stacking). Support vector machines and neural networks. Introduction to methods and algorithms for deep learning. Model interpretability and explainability. Statistical analyses using R or Python. Pre: Graduate Standing.

Pathways: N/A

Course Hours: 3 credits

Corequisites: N/A

Crosslist: ADS-5525

Repeatability: N/A