Accurate porosity prediction in oil and gas reservoirs is crucial for reservoir characterization. However, traditional methods like core measurements and logging data are expensive, inefficient, and fail to fully capture large-scale 3D porosity distributions. To address the limitations of petrophysical models with specific assumptions, this paper introduces an Explainable Artificial Intelligence (XAI) approach that combines machine learning algorithms, SHapley Additive exPlanations (SHAP) analysis, and petrophysical theory. Using the F3 Block in the European North Sea Basin as a case study, we built 3D porosity prediction models with four mainstream algorithms—multilayer perceptron (MLP), Random Forest (RF), Support Vector Regression (SVR), and Linear Regression (LR)—using various 3D seismic attributes as inputs. SHAP values were employed to uncover the decision-making logic behind the predictions. All four algorithms achieved high accuracy, with support vector regression performing best (R²=0.9669, RMSE=0.0055), surpassing traditional linear regression. The seismic inversion acoustic impedance, given its clear petrophysical meaning, was the most influential input feature across all algorithms. Low-pass filtering of logging curves further improved prediction accuracy. This study proposes an efficient, reliable XAI method for porosity prediction by integrating machine learning, petrophysical theory, and SHAP analysis. This approach enhances model interpretability and transparency in reservoir characterization, ensuring results are petrophysically sound.