Learning analytics engine needs to be able to export models and send them back to HQ (via mnet?) so they can be combined into an ensemble model to improve predictive capability.

Examples:

Python scikit: http://machinelearningmastery.com/ensemble-machine-learning-algorithms-python-scikit-learn/

http://machinelearningmastery.com/machine-learning-ensembles-with-r/

This can make us less dependent on receiving full data sets from sites. We would be able to ship models validated by many sites on their own data, rather than having to do all of our data analysis in-house. Sites could share model data, as described in MDL-59059, without exposing any personal data, and without having to export and de-identify site data. We could also use the combined model to create a large synthetic data set for researchers.