Ivy Tech Project Early Success — Machine-Learning Early Warning for Failing Community-College Students, Indiana

In a fall 2016 pilot, Ivy Tech Community College analysed learning-management-system behaviour data across 10,000 course sections and flagged about 16,000 students statistically at risk of failing by the second week. Outreach workers phoned them over the following two weeks.
The vendor case study and SiliconANGLE report that early models predicted final course grades with 60–70% accuracy at week two, later around 80% after moving to Google Cloud and TensorFlow, and that 98% of contacted students were neutral or satisfied with the outreach. Reported results: D and F grades fell 3.3% in the first term, with about 3,000 students helped to reach a C or better.
Caveats: the outcome figures come from the institution and its technology vendor, and no baseline or control-group comparison or long-term tracking is documented. Nothing public describes bias auditing of the predictions, a material concern for a system profiling students at an open-access college.
Value: a practical, low-cost pattern of prediction plus human outreach; the evidence supports it as promising, not proven.

Read the full analysis: https://edu.google.com/intl/ALL_us/why-google/customer-stories/ivytech-gcp/