A data-driven Machine Learning approach to creativity and innovation techniques selection in solution development

The creation and refinement of new ideas is a strategic competence for teams and organization to innovate and prosper. This paper addresses the challenge of finding adequate creativity and innovation techniques (CITs) for improving individual or team creativity through the use of Machine Learning (ML). The process of choosing which CIT to use is complex and demanding, especially when taking into consideration the existence of hundreds of techniques and the plurality of different design contexts. This empiric knowledge, usually retained in an expert's repertoire, can be extracted and implemented in a computational system, making it more available and permanent. This research focused on developing a Decision Support System embedded in an online application with a two-stage ML inference process able to evaluate users' design scenario through an online form, and infer the most appropriate CITs from the database that would fit their needs. This paper presents two iterative development cycles of the prototype, first focused on core knowledge acquisition, representation, ML implementation, and verification; while second focused on system expansion, addition of web interface, and initial validation. After essaying 12 algorithms, the two-stage model achieved uses a Gradient Boosted Regression Trees algorithm using user provided information about the context to infer the required CITs characteristics; followed by a Logistic Regression classification-ranking algorithm that uses outputs from first model to define which CITs to present to users. To the best of our efforts, no other system was found to use ML approaches to address the problem of CIT selection. (c) 2022 Elsevier B.V. All rights reserved.