Datarobot pioneers automated machine learning
Founded in 2012 in the United States, DataRobot is a pioneer in automated machine learning (Auto ML). Based on a problem to be solved (a financial prediction, preventive maintenance, image recognition), its AutoML system selects and trains several possible algorithms, before scoring them. This allows the user to deploy only the best performing model in production.
Detecting cardiac anomalies with AutoML
The following video describes the training of 19 models by Datarobot with AutoML on a problem of classification of patients likely to have cardiac anomalies or not.
The best performing model is then deployed in production to make predictions on a test data set.
- eXtreme Gradient Boosted Trees Classifier with Early Stopping is the best model identified by AutoML.
- Deployment of the predictive model in production to make predictions.
- Predictions made with a web interface provided by DataRobot
Use of Datarobot by healthcare actors
Datarobot proposes models of problems that are well known to healthcare actors: reducing the length of stay of patients, reducing the readmission rate to 30 days, predicting excessively high medical claims (fraud, waste, abuse).
Datarobot Launches ContagionNet to Combat Covid-19 and Future Pandemics
During the covid-19 pandemic, modeling research has shown that, of those infected, approximately 20% with the highest viral load cause 80% of new infections. On average, people who test positive for COVID-19 receive their results 7 to 10 days after initial exposure.
Peak viral load — the period when people are most infectious — occurs within 3 to 5 days of exposure, and new patient studies confirm that people with the highest viral loads are the most infectious.
These studies show that by the time people receive their test results, they have likely infected their closest contacts and beyond.
The goal of ContagionNET, a nonprofit subsidiary of DataRobot launched in February 2021, is to identify people earlier than traditional testing ( when they are most contagious) in order to disrupt the chain of transmission.
ContagionNET provides a suite of tools, including at-home antigen testing and a digital platform, to help identify the most contagious areas of a community and encourage people to take preventive measures.
