ML #28 - 2.0

Hosted by Mike Mastanduno

May 03, 2018 - 30min

Share this content:

We are excited to announce the next generation of our healthcare machine learning platform: v2.0!


Join us in this broadcast for a walkthrough of how to use the new package to easily make predictions using a highly refined model that is customized to your data.


The new R package integrates lessons learned from installations at over 15 health systems and brings substantial machine learning power to even the novice user. Highlights include automatic data preparation, feature engineering, algorithm selection, and hyperparameter tuning, as well as headache-free model deployment. The advanced user can customize a wide variety of parameters, while for beginners, or for those who just want machine learning predictions quickly, you can simply run:


models <- machine_learn(patient_data, outcome = readmission) predict(models, new_patient_data) will automatically fix common problems such as missing data and near-zero variance columns, engineer features such as day-of-week and hour-of-day from timestamps, and tune multiple predictive models using cross-validation to optimize performance. When making predictions, remembers any data manipulation that was performed in model training so that prediction datasets are always prepared identically to the training dataset, and makes predictions using the model and specifications that maximize predictive power.

What topic or projects should we feature?

Let us know what you think would make it great.