High-resolution lithofacies prediction using CWT-enhanced seismic attributes and gradient boosting: A case study from the Lower Goru Formation

Lithofacies prediction in complex clastic reservoirs demands workflows that can resolve thin-bed variability while minimizing interpreter bias. In the Lower Goru Formation, a machine learning framework was implemented to predict key petrophysical properties; Volume of Shale, Effective Porosity, and Water Saturation using integrated well log and seismic datasets. Two machine learning algorithms, Gradient Boosting Regressor (GBR) and Feed Forward Neural Network (FNN), were calibrated on both conventional seismic attributes and Continuous Wavelet Transform (CWT)-enhanced inputs. Enhanced vertical resolution from the wavelet-domain representation significantly improved prediction accuracy and continuity of lithologic patterns across the seismic volume. On the CWT-enhanced dataset, GBR achieved R² values of 0.99 for Volume of Shale and Effective Porosity, and 0.97 for Water Saturation, outperforming FNN while maintaining computational efficiency. The predicted lithofacies map delineated laterally continuous gas sand bodies in the northeastern and southeastern sectors, bounded by wet sands and overlying shale. Reservoir heterogeneity, previously unresolved in conventional workflows, was successfully captured. The results demonstrate a robust, data-driven approach to high-resolution lithofacies classification suitable for both development and frontier exploration environments.