MLMorph - Machine Learning-based Morphological Modeling
Below are two sections: 1) description of the MLMorph method, and 2) live online tool, which enables you to create a neural network and subsequently a morphological model based upon it.
About the MLMorph
This serves as a very brief introduction to morphological modeling. For a more detailed overview, refer to the works of F. Zwicky and T. Ritchey. Morphological modeling involves two key features: 1) creation of the solution space, and 2) reduction of this space to obtain workable solutions appropriate for the research or practical objectives. By leveraging combinatorics, one can explore appealing solutions to various scientific and practical problems.
The steps involved in morphological modeling are as follows:
- Problem formulation
- Identification of solution parameters (variables and their levels)
- Creation of morphological space (all possible solutions based on combinatorics)
- Reduction of morphological space (limiting to attractive solutions)
- Evaluation of the remaining solutions (e.g., through experiments or simulations)
- Selection of the best solution
The morphological method (basis for MLMorph), in its current form, was introduced and popularized by F. Zwicky, an American astrophysicist of Swiss origin. According to Zwicky, the first significant project using this method began in 1939 at the Guggenheim Aeronautics Laboratory at the California Institute of Technology. The project focused on developing rocket propulsion systems, which led to the formation of the Aerojet Engineering Corporation in 1942. Zwicky, as Director of Research, analyzed 576 combinations of propulsion systems, yielding surprisingly successful results (Zwicky, 1967a, 1967b, 1969).
The promotion and development of this method are carried out by the Morphological Society of Zurich (Morphologische Gesellschaft Zürich) and the Fritz Zwicky Foundation (Fritz‑Zwicky‑Stiftung). Recently, the Swedish Morphological Society, particularly T. Ritchey, has advanced morphological analysis into a general scientific modeling method (Ritchey, 2018, 2022; Ritchey & Arciszewski, 2018).
Historically, morphological modeling relied on a deductive process. Zwicky’s approach followed a hypothetico-deductive method, with the creation and reduction of morphological space guided by assumptions and evaluated by experiments. Ritchey’s method is mainly deductive, relying on an a priori basis as a fundamental scientific modeling method. While robust, Zwicky’s approach can be resource-intensive, and Ritchey’s is primarily theoretical unless supported by empirical validation.
MLMorph introduces a deductive-reductive process. Problem formulation, identification of solution parameters, and creation of morphological space are done deductively, as in Zwicky’s and Ritchey’s approaches. However, the reduction of morphological space and evaluation of solutions are conducted reductively (to be specific, inductive reasoning is employed). This inductive part involves training a machine learning model (e.g., neural network, random forest) to simulate outputs for all combinations, reducing and evaluating possible solutions, and supporting creative solutions not previously observed. While this approach is susceptible to issues arising from inductive reasoning, it offers a robust method to analyze empirical data quickly and propose decision-focused morphological models using complex machine learning simulations. Additionally, the software allows for simple representation of complex models, including morphological models.
MLMorph is not limited to simplex models; it can also deliver duplex models. Machine learning can handle both categorical and numeric variables, which can be binned for subsequent morphological space creation. Additionally, this approach offers the possibility to add open-ended content, like text (e.g., mission statements or strategy documents), images (e.g., photos of a product), or even recordings to shape the morphological space according to such input. For example, a morphological duplex model might be created that takes product photos and descriptions as part of the configuration and provides information on the best combination of marketing-mix (4P) with regard to sales. The demo on this site, due to technological reasons (TensorFlow.js is used), is restricted and does not allow for the creation of such models. Please contact me if you are interested in consulting services or academic research.
References
- Ritchey, T. (2018). General morphological analysis as a basic scientific modelling method. Technological Forecasting and Social Change, 126, 81–91.
- Ritchey, T. (2022). Conceptual Modelling, Combinatorial Heuristics and Ars Inveniendi: Ramon Llull and the Combinatorial Art An Epistemological History. Acta Morphologica Generalis, szkic.
- Ritchey, T., i Arciszewski, T. (2018). Editors’ introduction. Technological Forecasting and Social Change, 126, 76–80.
- Zwicky, Fritz. (1967a). Morphology of multilanguage teaching. In: F Zwicky i A. G. Wilson (Red.), New Methods of Thought and Procedure (314–331). Berlin: Springer-Verlag.
- Zwicky, Fritz. (1967b). The Morphological Approach to Discovery, Invention, Research and Construction. W: F Zwicky i A. G. Wilson (Red.), New Methods of Thought and Procedure (273–297). Berlin: Springer-Verlag.
- Zwicky, Fritz. (1969). Discovery, Invention, Research Through the Morphological Approach. Toronto: The Macmillan Company.