Credit-Delinquency Modelling and the Financial-Cooperative Gap: A Quantitative Review of Determinants, Ensemble Learning and Explainable Artificial Intelligence

Credit delinquency is the central prudential risk of savings and credit cooperatives, yet the methods used to model it have been developed and tested mostly on banks, consumer credit and benchmark datasets. This review uses the broader credit-delinquency literature as a benchmark against which the position of financial cooperatives is measured. A curated corpus of 123 documents (1953–2026), assembled by expert selection and citation tracing and verified against bibliographic registries, was analysed with bibliometric laws, natural language processing (TF–IDF, non-negative matrix factorisation, co-word networks, semantic drift and burst detection), a binomial generalised linear model of paradigm diffusion based on manually verified coding, a citation-free contribution index and a contingency test of coverage deficit. Five phases are identified, with the largest semantic displacements at the 1997 and 2016 boundaries (cosine drift 0.642 and 0.567). Explainability diffuses fastest (odds ratio 1.354 per year, 95% CI 1.202–1.526), and the mean number of paradigms per document rises from 0.91 to 2.36. Cooperative research follows a different path: none of the 12 cooperative documents employs ensemble learning or explainability methods (Fisher exact p = 0.0001), and a reproducible search of 65,610 Semantic Scholar records indicates that cooperative credit-risk studies apply these methods at roughly one quarter of the rate observed elsewhere (0.8% versus 3.0%). The periodisation, the rapid diffusion of explainability and the cooperative coverage deficit hold when the incomplete year 2026 is excluded and under alternative filter thresholds, topic numbers, network and burst thresholds and coding procedures. A conceptual five-layer framework and a research agenda are proposed to guide empirical work on explainable delinquency modelling in cooperatives.