Meta-learning approach for implementation of AI methods in the context of CRISP-DM with case studies from master data management
Victor Pontello, H. T. Beckmann, Carsten Lanquillon · 2021
Implementing artificial intelligence (AI) methods is often not a trivial task. Especially for beginners, it is a big challenge to find out a suitable AI method to a certain problem. Moreover, a heuristic approach to algorithm selection is often error-prone and suboptimal. As a solution to this problem, a process model was developed on the basis of a meta-learning approach in the context of CRISP-DM. The model uses a criteria grid to characterize the problem and, based on these characteristics, the process model is able to propose optimal AI methods as possible solutions. The model was evaluated using tasks from master data management and the observed results find support and recognition in the literature.