-
Notifications
You must be signed in to change notification settings - Fork 2
Home
Welcome to the Linnaeus Classifier wiki!
During internal discussions, it became apparent that the primary off-ramp for Auto-CEST is the free text field on 526EZ forms. Veterans write in their contentions but usually fail to use standardized language to identify them. With the high number of contentions on the average claim, it is statistically unlikely that all contentions will correspond perfectly to the list of contentions that are predefined in D2D Web Services (Direct Link to Excel File). All claims that fail auto-CEST are reviewed by a Claims Assistant and the work is manually completed. It was identified that fixing this defect would increase the Auto-CEST rate from roughly 1% of claims to approximately 40% of claims. This would reduce workload in the field significantly and potentially free up hundreds of FTE to work on more value-added activities. Furthermore, this application would reduce the VBA’s Average Days Pending (ADP) for compensation claims by approximately 5 days, meaning that Veterans would be paid a week earlier on average
After understanding the business need, two AI technologies were identified that solve this problem. Natural Language Processing (NLP) and Machine Learning (ML) could be combined using a Python code in order to use the hundreds of thousands of lines of free text to train an AI to automatically classify contentions with an 80 to 90% accuracy. This code is the working model that hopefully will increase the service provided to all Veterans by the VA.
The Linnaeus Classifier is named for Carl Linnaeus who is recognized as the father of modern taxonomy. The goal of this small application is to automatically classify data in accordance with the model that is built using machine learning within the Model Builder. Since this is the deployable application, all documentation and project tracking will be done here.
This project wouldn't have happened without Alex and his tutorial on Python, the scikit-learn, and using pandas to create machine learning models. Thanks Alex for all your help!