<p>Precise characterization of reservoir rocks, particularly regarding porous media properties such as porosity, pore throat permeability, and fluid saturation, is essential for efficient hydrocarbon extraction and management. Traditionally, these properties have been assessed through core sampling and well log analysis. However, the data obtained from point-by-point measurements using these methods are often not generalizable to the entire reservoir's porous media due to the inherent heterogeneity of reservoir rocks, spatial variability, and limited sampling intervals, resulting in significant uncertainty in extrapolation. Recent advancements in data-driven techniques offer promising solutions to overcome these limitations, enhancing the predictive accuracy and interpretive power of petrophysical data. This study investigated the application of leading deep neural network algorithms to model the complex relationships between petrophysical characteristics and porous media properties derived from core samples. Using a dataset comprising 3549 records from three wells in a Middle Eastern oilfield, the research demonstrated the effectiveness of long short-term memory (LSTM) models in capturing nonlinear patterns often overlooked by traditional methods. Principal components analysis (PCA) was used for feature reduction, highlighting key parameters such as medium resistivity (RES-MED), compressional-wave velocity (<i>V</i>p), and the reservoir quality index (RQI) as significant factors influencing reservoir quality. The LSTM model outperformed conventional models, achieving exceptional accuracy with MAE = 0.0001, RMSE = 0.0091, and <i>R</i>² = 0.9856. These findings underscore the potential of machine learning/deep learning models to reduce reliance on labor-intensive core sampling, streamline reservoir characterization, and provide more efficient, cost-effective methodologies for evaluating reservoir quality and optimizing hydrocarbon recovery.</p>