Machine Learning: Theory and Applications

Our goal is to provide a clear and accessible “behind the scenes” understanding of many machine learning procedures for prediction and classification that are more frequently being used in the health and behavioral sciences. We strive to strike a balance between core concepts of the algorithmic procedures along with their application and interpretation when implemented with real empirical data. This workshop is designed to provide participants with the knowledge to discern which machine learning models are appropriate for their research question and the subsequent tools to analyze their data in ways not previously possible.

This workshop is designed for graduate students, post-doctoral fellows, faculty, and research scientists from the behavioral, social, and health sciences. Although it is recommended that participants have a working knowledge of the general regression model, these core concepts will be reviewed at the beginning of the workshop. Participants who seek a more in depth review prior to attending may wish to watch our Linear Regression Playlist on YouTube.

Chapter 1: General Introduction
1.1 Introduction to the Predictive Modeling Process
1.2 Matrix Algebra Review
1.3 Data Pre-processing
1.4 Principal Component Analysis

Chapter 2: Over-Fitting and Model Tuning
2.1 The Problem of Over-Fitting
2.2 Model Tuning and Data Splitting
2.3 Resampling Techniques
2.4 Choosing Between Models

Chapter 3: Regression Models
3.1 Variance-Bias Trade-Off
3.2 Linear Regression
3.3 Partial Least Squares
3.4 Penalized Models/Regularization
3.5 Multivariate Adaptive Regression Splines

Chapter 4: Regression Trees
4.1 Regression Tree Models
4.2 Rule-Based Models
4.3 Model Aggregation: Bagging/Boosting
4.4 Random Forests

Chapter 5: Linear Classification Models
5.1 Evaluating Classification Models
5.2 Logistic Regression
5.3 Linear Discriminant Analysis and Variants
5.4 Nearest Shrunken Centroids

Chapter 6: Nonlinear Classification Models
6.1 Nonlinear Discriminant Analysis
6.2 Flexible Discriminant Analysis
6.3 Support vector machines

Chapter 7: Classification Trees
7.1 Classification Tree Models
7.2 Model Aggregation: Bagging/Boosting
7.3 Random Forests
7.4 C5.0 Algorithm

Chapter 8: Additional Considerations
9.1 Addressing Class Imbalance
9.2 Variable (Feature) Importance and Selection
9.3 Type III Errors (rejecting the null for the wrong reasons)
9.4 Measurement Error
9.5 Trusting Your Model

Livestream lectures will begin at 10:00 a.m. and finish at approximately 3:30 p.m. Eastern Time (US) each day. There will be 10-minute morning and afternoon breaks and a one-hour lunch break, the exact times of which are determined during lecture. Local time zones within which the participant is connected must be adjusted to correspond to Eastern Time (US).

Pre-recorded software demonstrations for machine learning are provided for self-paced learning using R and Python. Note that both R and Python can be downloaded for free (R download; Python download) While it is helpful to have some familiarity with R or Python, it is not necessary. The lectures which constitute the majority of the workshop are software-independent.

Doug Steinley teaches the workshop live via Zoom, with recordings posted at the end of each day for those who might miss or wish to review any of the material. The lectures for the workshop are software-independent but demonstrations are provided in R and Python via pre-recorded videos.

Participants receive access to the livestream, one-year of access to recorded lecture and demonstration videos, downloadable complete PDF copies of the course notes and software demonstrations, as well as data and code for all examples. The PDFs are not time-limited and may be retained indefinitely but should not be distributed to others without obtaining prior permission.