Improving petroleum exploration and development is important worldwide for reducing drilling risks and understanding reservoir characteristics. This study tackles these challenges by presenting new methods that use machine learning algorithms to analyze complex seismic and well-log data in Egypt’s West Nile Delta Deep Marine (WNDDM) concession. The main goal of this study is to enhance reservoir rock characterization by integrating information from seismic attributes and well logs. To achieve this, a novel approach combining fluid substitution and machine learning techniques is used to assess how pore fluids affect reservoir elastic properties. Initially, gas-bearing zones were excluded to minimize the negative impact of gas on Poisson’s ratio predictions across three different machine learning models. These gas zones were replaced with oil and water equivalents, resulting in a more accurate prediction process that demonstrated improved results, validated by higher correlation coefficients. This thorough method allowed for a detailed examination of how gas, oil, and water affect Poisson’s ratio estimation. The study employed the Random Vector Functional Link (RVFL) model, fine-tuned using the Cheetah Optimizer (CO) and Wild Geese Algorithm (WGA), to investigate how pore fluid saturation affects reservoir elastic properties. The findings emphasized the significant impact of gas-bearing zones on the accuracy of Poisson’s ratio predictions, showing notable improvement when gas was replaced with water rather than oil. Regression analysis validated these results, revealing better correlation coefficients for the CO-RVFL and WGA-RVFL models after excluding gas-bearing well-log data and further enhancing when gas zones were substituted with oil or water equivalents. Furthermore, the study utilized seismic data and well logs to predict shale volume and effective porosity through the Random Forest (RF) algorithm. The results demonstrated the efficiency of this machine learning approach in estimating vital petrophysical parameters, underscoring the potential of artificial neural networks in advancing reservoir characterization methods.