It's easy to fool yourself: Case studies on identifying bias and confounding in bio-medical datasets
This highlights critical issues for researchers using ML on raw biomedical data, though it is incremental as it builds on known challenges in confounding.
The paper tackles the problem of hidden biases and confounding variables in biomedical datasets when using black-box machine learning, revealing through two case studies that models performed well but were flawed due to systematic errors and spurious signals.
Confounding variables are a well known source of nuisance in biomedical studies. They present an even greater challenge when we combine them with black-box machine learning techniques that operate on raw data. This work presents two case studies. In one, we discovered biases arising from systematic errors in the data generation process. In the other, we found a spurious source of signal unrelated to the prediction task at hand. In both cases, our prediction models performed well but under careful examination hidden confounders and biases were revealed. These are cautionary tales on the limits of using machine learning techniques on raw data from scientific experiments.