When building predictive models, model accuracy, measured by metrics like precision, recall and area under the curve (AUC), has traditionally been the primary driver of model design and operationalization. While this leads to high-fidelity model construction at training and testing time, performance in production often degrades, producing results that are worse than expected.
As machine learning (ML) matures within organizations, resiliency often overrides raw predictive accuracy as the defining criterion for operationalizing models. Increasingly, ML practitioners are leaning towards operationalizing well performing, predictable production models rather than those that exhibit high performance at testing but don’t quite deliver on that promise when deployed.
Read the Full Article on Towards Data Science →
Towards Data Science Inc. using Medium, provides a platform for thousands of people to exchange ideas and to expand our understanding of data science. Our audience is mixed, consisting of readers entirely new to the subject and expert professionals who want to share their inventions and discoveries.