Machine-Learning-Assisted Mud-Gas Geochemistry for Reservoir-Fluid Affinity and Drill Bit Metamorphism Assessment

Real-time mud-gas geochemistry provides early insight into reservoir-fluid composition during exploratory drilling, but interpretation can be affected by mud properties, drilling conditions, degassing efficiency, and Drill Bit Metamorphism (DBM), a thermomechanical process that can generate artificial unsaturated hydrocarbons and distort native fluid signatures. This study presents a machine learning-assisted workflow to quantify reservoir-fluid compositional affinity and drilling-induced geochemical alteration using an integrated dataset from 104 Brazilian exploration wells. The workflow combines conventional and Advanced Gas (AG) measurements, drilling parameters, mud properties, bit specifications, laboratory pressure-volume-temperature (PVT) fluid compositions, and expert geological interpretation. After quality control and exploratory analysis, depth dependent compositional similarity between AG measurements and PVT fluid compositions was used to construct C2- and C2C-based Reservoir Affinity Curves, whereas ethylene-related behavior and operational variables were used to generate a DBM Severity Curve. Joint interpretation of C2, which includes ethane and ethylene, and C2C, which isolates ethane, provided an additional geochemical criterion for recognizing DBM-related ethylene enrichment. Kernel Ridge Regression (KRR) was selected as the primary deployment model because it produced stable, smooth, and interpretable depth profiles, while XGBoost and LightGBM served as accuracy benchmarks. KRR achieved R² values of 0.84 for C2 affinity, 0.83 for C2C affinity, and 0.84 for DBM severity, whereas gradient-boosting models achieved R² values of up to 0.98 for affinity and 0.91 for DBM severity. Gas-composition variables dominated the affinity models, whereas mud properties and bit-related operational variables were more influential in DBM prediction. High affinity in both C2- and C2C-based curves was associated with more representative formation-fluid signatures, whereas elevated ethylene coupled with reduced affinity identified intervals requiring greater caution in geochemical interpretation. The proposed approach transforms mud gas geochemical information into quantitative, traceable indicators that support earlier fluid characterization, improved sampling decisions, and reduced uncertainty during exploratory drilling.