The pore structure in carbonate rocks is intricate and heterogeneous, containing intraparticle and interparticle porosities. Assessment of hydrocarbon recovery in carbonate reservoirs without considering the presence of microporosity may lead to inaccurate estimation of hydrocarbon recovery factors. Conventional well-logging techniques struggle to measure microporosity in such complex lithologies. As nuclear magnetic resonance (NMR) is independent of the rock matrix and mineralogy, it becomes the preferred method to measure microporosity, but it is expensive and not widely available. This study proposes an integrated approach utilizing supervised machine learning (ML) techniques to enhance the prediction of NMR-based porosity using various well logs as input features. The study utilized 6000 data points gathered from multiple wells located in a carbonate reservoir to predict NMR-based macro- and microporosities as a function of corresponding well logs using various techniques, including decision trees, K-nearest neighbors, gradient boosting, random forests, adaptive gradient boosting, and multilayer dense neural networks. In this study, we categorized subsurface rocks into distinct clusters/facies using measurements obtained from well logging. This clustering allowed us to establish relationships between these clusters/facies and the predicted porosities of the rocks. Ultimately, the prediction performance of the models was evaluated using the coefficient of determination (R²) and mean absolute percentage error (MAPE). The results indicate that an integrated approach can effectively predict porosity based on NMR data. The dense neural network emerged as the top-performing supervised learning algorithm, achieving an impressive R² value of 0.98 and a MAPE of 3%. The proposed model for NMR porosity prediction relies solely on the quantity of hydrogen nuclei in pore spaces, thereby eliminating the influence of rock minerals and focusing exclusively on the pore.