Typology and taxonomy of monolithic and hybrid AI/ML predictive models in finance: an integrated generational architecture

Abstract

The rapid proliferation of artificial intelligence and machine learning models applied to financial forecasting has produced a deeply fragmented analytical landscape, where heterogeneous architectures coexist without any shared classificatory framework to systematically organize or compare them. To address this gap, the article develops a structured nomenclature grounded in a dual logic: a generational progression and a multi-dimensional articulation, encompassing both monolithic and hybrid configurations deployed in finance. The methodology is simultaneously deductive and inductive, anchored in the iterative protocol introduced by Nickerson, Varshney, and Muntermann (2013), and aligned with the FAIR principles (Wilkinson et al., 2016) and the taxonomic validity criteria of Bailey (1994) and Doty and Glick (1994). Conceptual validation rests on twelve illustrative cases, one per class, selected through triangulation. The resulting framework revolves around three primary dimensions (Architecture, Family, Generation) yielding twelve classes: nine monolithic configurations and three hybrid classes. The framework constitutes an integrated and reproducible contribution to the comparative mapping of predictive AI/ML models in finance.

Classification JEL: C45, C53, G17

Paper type: Theoretical Research