This presentation by Eric Munger, Ph.D., explores the application of machine learning and deep learning
techniques, specifically focusing on a traumatic brain injury (TBI) study using the VA TBI Model Systems
(TBIMS) database. It contrasts traditional machine learning with deep learning, detailing various algorithms
like random forests and neural networks. The presentation emphasizes a Random Survival Forest model used to
predict patient retention in the TBIMS study, analyzing variable importance and model performance using metrics
such as the Brier score. Additionally, it introduces Generative Adversarial Networks (GANs) as a method for
generating synthetic data. The overall goal is to demonstrate the utility of AI in analyzing complex TBI datasets
and improving predictive modeling.
Resources
The white paper below, Applicability of Machine Learning to the TBIMS NDB,
describes three standard machine learning algorithms as applied to the TBIMS NDB: Support Vector Machine,
Artificial Neural Network, and Random Forest. Additionally, there are R packages that researchers can download
and use to clean the TBIMS NDB data (TBIMSDatacCleaning_0.1.0.tar.gz) and run the code for the three algorithms
(TBIMSMachineLearning_0.1.0.tar.gz). Questions should be directed to the NDSC at tbimsdata@craighospital.org.>