Humana-Mays Data Science Competition

Predicting Medicare Members Most Likely Struggling with Transportation Issues

Authors: Brian Bacik, Chiu-Feng Yap

Project Status: [Completed]

Project Ranking: Top 50/276

Introduction

Using 1-year worth of retrospective data of Humana MAPD members, competition participants are required to analyze the data to develop a predictive model that takes in a set of features of customers and assigns to each member a probability of being at risk for a transportation challenge. These probabilities will be evaluated by the ROC AUC paradigm, essentially measuring the ability of the model to correctly classify members.

Methods Used:

• Categorical variable encoding
• Feature engineering
• Hyperparameter tuning
• Dimension reduction

Models Used:

• Logistic Regression Classifier
• Lasso Regression
• Elastic Net
• Support Vector Machine
• Random Forest
• XGBoost Classifier

Data Source

Raw training data and the holdout set are provided by Humana. Data comes from various sources documenting members’ medical-related behaviors, including:

• Medical claims
• Pharmacy claims
• Lab claims
• Demographic/consumer data
• Credit card
• Clinical condition related
• CMS member data elements

Target Variable

The target variable (transportation issue) is a self-reported binary variable (“0” - No, “1” - Yes) to indicate transportation challenges. Transportation screening question:

“In the past 12 months, has a lack of reliable transportation kept you from medical appointments, meetings, work, or getting things needed for daily living?”

is coming from the Accountable Health Communities - Health-Related Social Needs Screening Tool. The survey was completed in November/December 2019.