Predicting melanoma risk from electronic health record data

Presentation by Aaron Richter (https://www.linkedin.com/in/aaron-richter-8b13437b):

This talk will present a case study using electronic health record (EHR) data to predict individual patient risk of developing melanoma in the future. The goal is to outline an example of a data science problem from start to finish. We will focus on:

• Why do we need a predictive model anyway?
• Dataset pre-processing (converting longitudinal patient data into a sparse matrix)
• Modeling: linear models, tree models, class imbalance
• Evaluating models: sensitivity, specificity, AUC
• Interpretability considerations: global vs. local