Intelligent system for automatically testing and selecting from multiple data models for accurate diversion prediction

Abstract

Methods, systems, and computer-readable media are disclosed that the likelihood that a medical order for multiple pharmaceutical drugs may be stolen or diverted. Generally, a current data set for a medical order for pharmaceutical drugs is received. Each of the drugs is associated with a set of features. An impact score is generated for each of the features for each drug based on historical effects. Test data is also used to evaluate the diversion prediction accuracy of a plurality of machine learning models, when compared to historical diversion data for the drugs. The most accurate machine learning model is utilized to make a diversion probability prediction for those features having the highest impact scores, for the drugs in the medical order. A recommended action is generated and provided based on the diversion probability.