Introduction to Machine Learning and Risk Prediction

EPI 288 Subject & Catalog Number

Description

Introduces machine learning and risk prediction methods with applications to clinical data. The basic concepts of supervised and unsupervised machine learning methods, as well as appropriate applications will be discussed. Topics covered will include:

• regularized regression techniques: ridge and lasso regressions
• tree-based algorithms: random forests and boosted trees (XGBoost)
• deep learning methods: including convolutional neural networks
• unsupervised methods: t-SNE and UMAP
• additional methods as time permits

 
We will also discuss key concepts for training prognostic and diagnostic models, such as cross-validation, adjustments for multiple comparisons, overfitting, and model performance metrics.

Students may complete homework assignments using either the Windows-based JMP programming language or R. Those choosing R are expected to have proficiency at the intermediate level.

Meeting Note: This is an online course with pre-recorded lectures; live online office hours, and seminars that will also be recorded and shared with the class. Therefore, this course will not have specific lecture meeting times.
 
Prerequisite: To enroll, students must have received a grade in one of the following classes: EPI 522, EPI 236, BST 210, BST 213, or BST 215. Or permission of the instructor.