QUA³CK - A Machine Learning Development Process
July 21, 2020
Machine learning and data processing are trending topics at the moment. However, there is still alack of a standard process to support a fast, simple, and effective development of machine learningmodels for academia and industry combined. Processes such as KDD or CRISP-DM are highlyspecialized in data mining and business cases. Therefore, engineers often refer to individualapproaches to solve a machine learning problem. Especially in teaching, the lack of a standardprocess is a challenge. Students typically get a better understanding if a systematic approach tosolve problems is given to them. A challenge when formulating a machine learning developmentprocess is to provide standard actions that work on different use-cases. At the same time, it has tobe simple. Complex processes often lead to the wrong approach.The QUA³CK process was created at the Karlsruhe Institute of Technology to fill the gap inresearch and industry for a machine learning development process. However, the main focus wasto reach engineering students with an easy-to-remember, didactic way to solve machine learningproblems. This five-stage process starts with a machine learning question (Q), a problem thathas to be solved. Understanding the data (U) comes next. Then, the loop between selecting anAlgorithm (A), Adapting the features (A), and Adjusting the hyperparameters (A) is executeduntil the system is ready for Conclude and compare (C). At last, the Knowledge transfer (K) ofthe given solution can be realized as deployment in hardware or as a documentation.This paper describes the process and all individual steps in detail. Besides, we present severaluse-cases of QUA³CK in academia and research projects.
How to cite
Metadata are provided both in "article" format (very similar to INSPIRE) as this helps creating
very compact bibliographies which can be beneficial to authors and
readers, and in "proceeding" format
which is more detailed and complete.