Integrating multi-proxy geochemistry and machine learning to decode cretaceous palaeoenvironments and tectonic evolution in the Nigerian Middle Benue Trough, West Africa.

The Middle Benue Trough, a key segment of the West African Rift, preserves a complex Cretaceous stratigraphy important for understanding basin evolution and hydrocarbon potential. We integrate sedimentological (foraminifera, palynofacies, clay minerals) and geochemical (major oxides, trace elements, REEs including weathering indices such as the chemical index of alteration) proxies in a machine-learning framework. To mitigate overfitting on our modest sample set, we preselected ~ 15 key proxies for analysis. Unsupervised learning (PCA, hierarchical clustering, t-SNE) clearly partitioned the Albian Asu River and Coniacian Awgu shale samples. Geologically, the Asu facies is interpreted as a restricted hypersaline lagoon (high Sr/Cu; abundant agglutinated foraminifera), whereas the Awgu facies records an open-shelf transgression (lower Sr/Cu; higher V/Cr, U/Th; organic-rich shales). These contrasts reflect a warm, humid depositional climate. A Random Forest classifier, evaluated by repeated stratified five-fold cross-validation, achieved ≈ 96% mean accuracy (AUC ≈ 0.98). Feature ranking in RF highlighted the Eu/Eu* anomaly and LREE/HREE ratio among the top predictors, indicating subtle REE signals can be as diagnostic as classical proxies. Other classifiers (SVM, XGBoost, kNN) gave consistent results. We emphasize that these ML analyses provide independent quantitative supports for the interpretation of the results of conventional proxies. Importantly, the models largely supported the conventional Asu/Awgu facies grouping rather than unveiling new depositional types. Bulk geochemistry (CIA ≈ 80-90) and tectonic indicators imply deposition on a rifted passive margin under a humid-to-semiarid climate. This integrated ML approach refines the Benue Trough stratigraphic framework and provides a reproducible, data-driven basis for paleoenvironmental reconstruction and exploration-risk assessment in similar rift basins.