Driving impact without an "answer": Unsupervised provider solutions in healthcare

The Provider Insights Data Science team at HCSC improves healthcare outcomes through surfacing data-driven provider insights. This talk highlights two anomaly detection projects leveraging unsupervised learning: provider working specialty attribution and anomalous opioid prescriber model. The provider working specialty solution attributes specialties to providers leveraging TF-IDF. This solution is important because it improves the results of downstream unsupervised solutions where no "answer" is available. The second talk focuses on an anomaly detection model leveraging an isolation forest to surface providers with unique behavior patterns compared to peers. Talk also features an overview of data governance and responsible AI.