Novel hybrid SOM-MRGC clustering with SHAP-optimized machine learning for integrated reservoir characterization: A case study of the Kazhdumi Formation

• Novel SOM-MRGC hybrid achieves 93% electrofacies accuracy (+40% vs baselines) in heterogeneous reservoirs. • δ¹³C > +3‰ threshold quantifies dolomitization-induced 22% porosity enhancement. • SHAP-optimized ML reduces permeability error to 0.12 log(mD) (40% improvement). • Geomechanical profiling identifies high-potential "brittle" facies (BI=0.71) for optimized hydraulic fracturing. • Field validation demonstrates 28% NPT reduction and 2.1 × higher production in optimized wells. The accurate characterization of heterogeneous carbonate-siliciclastic reservoirs remains a critical challenge in petroleum geoscience due to complex pore architecture, lithological heterogeneity, and multi-phase diagenetic alterations. This study presents an innovative integrated workflow that significantly advances reservoir evaluation in the Kazhdumi Formation of southwest Iran by synergizing multi-resolution graph-based clustering (MRGC), hydraulic flow unit (HFU) analysis, and explainable machine learning. The novel hybrid SOM-MRGC clustering algorithm demonstrates superior resolution, identifying seven distinct electrofacies—a 40% improvement over conventional methods. Petrographic validation confirms these electrofacies correspond to specific diagenetic regimes. A quantitative link between dolomitization impacts, defined by a stable isotope threshold (δ¹³C > +3‰), and a 22% porosity enhancement is established, with Electrofacies 7 (EF7) exhibiting abundant fracture-vuggy porosity. The workflow demonstrates a direct electrofacies-HFU correlation (EF7=HFU4), confirming distinct flow capabilities. Furthermore, SHAP-optimized XGBoost models achieve superior permeability prediction (RMSE = 0.12 log(mD)), while geomechanical profiling identifies high-potential "brittle" facies ideal for hydraulic stimulation. The methodology's transferability and potential for broader application are supported by successful deployment in diverse reservoirs, including the Ordos Basin and Arab Formation, though basin-specific calibration is recommended. Field implementation demonstrates operational improvements, including a 28% reduction in non-productive time and higher initial production rates in optimized wells. This integrated approach establishes a new standard for reservoir characterization, and the proposed workflow offers a reliable and transferable approach for characterizing complex reservoirs, with direct implications for reducing operational risk and improving well placement.