KDD is the non-trivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data. This concept was formalized in the early 1990s, notably by Usama Fayyad and others, to distinguish between the full discovery process and the narrower step of applying data mining algorithms. Under this definition, data mining is actually a step within the broader umbrella of KDD. The concept of KDD is less commonly referenced today and in practice much of KDD has been absorbed into data science workflows and ML pipelines. KDD was born in an era of well-structured data stored in relational databases. The rise of [[deep learning]] has challenged the emphasis placed by KDD on "understandable" patterns, shifting more towards useful applications (whether interpretable or not).