Risk Modeling

Our team built, tested, and refined a predictive model, as proof of concept, that identified changes in beneficiary income that might impact eligibility. Two-Class Logistic Regression and Decision Forest models were built using Microsoft Azure Machine Learning and cross-validated for accuracy.

By analyzing historical data from multiple sources, and comparing against current data, the predictive model could determine the likelihood that income had changed and assign risk scores based on the client’s review criteria. The resulting scores were used to stratify and group cases, based on whether the cases required “No Review,” “Partial Review” or “Full Review.”

Our team demonstrated that its predictive model could accurately calculate risk scores for use in stratifying cases. This process helped to reduce the number of cases that required a manual review, allowing staff to spend more time on cases with greater potential for risk and/or using more senior staff to review high-risk cases. In addition, we demonstrated that the model could provide support for human decision-making and help to better manage the workloads of staff, making them more efficient and effective and, in some cases, automating the entire decision-making process.