Healthcare: Patient Recovery Rates


Researchers, hospitals, doctors, and healthcare insurers are applying machine learning tools to vast amounts of both clinical and non-clinical data to identify trends and issues that will improve patient care and outcomes. A major driver has been the implementation of electronic health records (EHRs) in many clinical and hospital settings, which has improved the usability and availability of data for machine learning purposes.

Machine learning lends itself well to many medical situations due to the binary nature of outcomes: was the patient readmitted after a specified period of time; did a particular condition reoccur; did the patient die? While the number of possible applications of machine learning in healthcare are vast, patient recovery rates in particular has significant impact not only to the healthcare industry, but to patients as well. Analyzing data on patient care regimes is allowing healthcare providers to better understand the factors involved in successful treatments, improving patient recovery rates. This, in turn, better predicts things such as who should be hospitalized and who can be treated in an outpatient setting, reducing the cost of providing healthcare to all involved.

Automated machine learning (AutoML) is particularly well suited to the task of analyzing the large datasets in the healthcare industry. But the process is often a long, drawn out one that can be very tedious. Auger’s AutoML tool greatly reduces the time spent in selecting algorithms and tuning hyperparameters by automating the process and trying every available algorithm and hyperparameter combination. This process generates a leaderboard showing the best combinations from which an ensemble is created, combining the best attributes of the leading algorithms and hyperparameter options.

Additionally, Auger employs a “warm start” process to algorithm/hyperparameter selection, generating more accurate results in less time.




Sign up and get free computing time!

  1. Upload your csv

  2. Train your models in parallel to find the best performer

  3. Deploy a prediction endpoint to get real time predictions