Comparative study of optimized rock-physics templates (RPTs) and machine learning (ML) approaches for sweet spot delineation in shale gas reservoir

Accurate prospectivity assessment for unconventional reservoir requires defining sweet spots, a methodology that delineates prospective areas based on quality factors such as organic richness, fracability, and maturity-related reservoir characteristics. In shale gas systems, critical parameters are typically defined by total organic carbon (TOC), brittleness index (BI), and gas saturation (Formula presented). This study compares sweet spot delineation using conventional rock-physics templates (RPTs) with machine learning (ML) algorithms, specifically Artificial Neural Networks (ANNs), Extreme Gradient Boosting (XGB), and Logistic Regression (LR). The RPT-based approach utilizes cross plots between different elastic parameters and employs a statistically supported objective function with two thresholding techniques—box-shaped and mathematical-function thresholds—where the former demonstrates superior performance. The ML approach leverages elastic parameters as features and applies a randomized search for hyperparameter optimization. Results show that the optimized ML-based approach is superior to the RPT-based one, achieving an (Formula presented) -score (obtained from precision and recall metrics) of 0.801 against 0.746. The analysis reveals that (Formula presented), (Formula presented), and (Formula presented) consistently rank as the most impactful elastic parameters based on validation performance across multiple feature combinations and ML algorithms for delineating prospective zones. Though less powerful, the RPT-based approach offers simplicity and may be optimized further or combined with the ML technique. Our findings underline the practicality and reliability of the proposed ML-based methodologies for unconventional reservoir assessment to accurately delineate sweet spots and improve reservoir evaluation practices.