Authorship in the Pauline Epistles: A stylometric and machine learning analysis (Extended Version)

Note on versioning: This deposit is an earlier, extended (long) version of the manuscript. A substantially revised and shortened version with a different title has been prepared for journal submission. This long version is shared for transparency and as supplementary background. The authorship of the New Testament epistles, especially those attributed to Paul, is a crucial issue for both biblical scholarship and contemporary theology. Although previous quantitative studies have offered valuable insights, their “black box” nature has limited their interpretive value for theological and historical scholars. This study adopts a systematic, data-driven, and transparent framework combining computational stylometry with interpretable machine learning to re-evaluate the authorship of the Pauline corpus. The methodological pipeline includes Okapi BM25 weighting, Principal Component Analysis (PCA), hierarchical and k-means clustering, and Linear Discriminant Analysis (LDA). A conservative classification model, trained exclusively on the core letters, recognized them with high accuracy. Crucially, explainable AI (SHAP) identified the specific, lower-variance stylistic components that are decisive for this classification, providing transparent, replicable evidence for what constitutes a core “Pauline” style. The findings quantitatively demonstrate that the Deutero-Pauline and Pastoral Epistles are stylistically different from this core. This integrated approach demonstrates how transparent data science can complement traditional exegesis, serving as a powerful bridge between quantitative analysis and historical theological inquiry. This extended version contains comprehensive methodological details, literature review, and exploratory analyses that serve as supplementary material for the primary manuscript.