Machine learning algorithms, increasingly used to map data into clinical predictions, are not as accurate as they should be, possibly making clinical study results that involve mobile tools inaccurate.
In fact, a new study at bioRxiv, describes the current state of machine learning as “voodoo,” and warns the erroneous results can prove misleading to clinicians and data scientists.
“As machine learning algorithms are increasingly used to support clinical decision making, it is important to reliably quantify their prediction accuracy,” the study's authors write.
They note that mHealth tool adoption is increasing the amount of data used in medical decision-making, but that erroneous results can mislead both clinicians and data scientists.
"As we move towards an era of machine learning based diagnosis and treatment, using proper methods to evaluate their accuracy is crucial," they add.
Mobile health app and device accuracy have been questioned before, mostly recently regarding apps geared toward fertility. A study, published in the Journal of the American Board of Family Medicine, found that just six of 40 evaluated fertility apps were accurate.
In addition, inMarch, a research letter reported that a once popular digital health app for blood pressure monitoring was highly inaccurate and providing three-quarters of users with incorrect readings.
For more information:
- here's a preview of the study