Published On Jun 28, 2024
“Yes” or “no” questions seem simple, but they can have profound consequences in healthcare. Is a patient portal message urgent? Will insurance deny this claim?
Machine learning can help us find the right answers. But building this kind of machine learning model requires making one of the trickiest decisions in model development: setting the threshold.
I review the process in my latest #CodetoCare video. Here’s the gist.
These machine learning models work by predicting the probability that an answer should be “yes.”
For example, a model might estimate an 81% probability that a specific message is urgent.
So the task then is to decide on the threshold which will then determine whether a message is identified is urgent or not.
Should the model flag all messages with a 70% or greater probability of being urgent? That sounds helpful — until you consider the potential for a flood of false positives.
Should developers set the threshold at 90%? The model might be more efficient, but it risks users missing important messages.
With no clear correct answer, determining the threshold is among the most challenging issues in ML today because it is essentially a business, workflow, and risk decision – not an ML decision.
Watch the video to learn my three rules for cracking this nut.
Check out my LinkedIn: / donwoodlock
---
ABOUT INTERSYSTEMS
Established in 1978, InterSystems Corporation is the leading provider of data technology for extremely critical data in healthcare, finance, and logistics. It’s cloud-first data platforms solve interoperability, speed, and scalability problems for large organizations around the globe.
InterSystems Corporation is ranked by Gartner, KLAS, Forrester and other industry analysts as the global leader in Data Access and Interoperability. InterSystems is the global market leader in Healthcare and Financial Services.
Website: https://www.intersystems.com/
Youtube: / @intersystemscorp
LinkedIn: / intersystems
Twitter: / intersystems
#AI #artificialintelligence #llm #genai #openai #gpt4 #ml #machinelearning #rag #generativeai #largelanguagemodels #largelanguagemodel #artificialintelligencetechnology #artificialintelligenceandmachinelearning
#healthcareIT #healthcaredata