Reservoir fluid identification is of paramount importance in petroleum exploration and development, directly impacting reserve assessment accuracy and drilling decision-making. However, conventional interpretation methods face persistent challenges including intrinsic non-uniqueness, strong dependency on prior knowledge, and limited resolution when dealing with complex reservoir systems. To overcome these limitations, this research develops an objective and efficient fluid identification methodology through the integration of multiple mud logging data types with advanced machine learning algorithms. The study was conducted in the Shahejie Formation of Qikou Sag, Bohai Bay Basin. We established a comprehensive dataset comprising 1539 validated samples with expert-verified fluid type labels (including oil, gas, water, dry, oil-water, and oil-gas zones). The research methodology comprised three main phases: First, we systematically analyzed the distribution patterns of three key mud logging data categories-gas logging (measuring C1-C5 hydrocarbons), quantitative fluorescence (characterizing aromatic components), and pyrolysis data (indicating free and bound hydrocarbons)-across different fluid types. Second, we implemented a systematic combinatorial optimization approach, generating seven input configurations from the three data categories and evaluating them with four distinct machine learning classifiers (K-Nearest Neighbors, Random Forest, Artificial Neural Network, and Support Vector Classifier). This resulted in 28 input-model combinations, whose hyperparameters were rigorously optimized using Bayesian optimization with Macro F1-score maximization as the objective function. Third, we employed the SHAP (SHapley Additive exPlanations) interpretability framework to elucidate the decision-making mechanism of the optimal model. The experimental results demonstrated that the Random Forest model with combined quantitative fluorescence and pyrolysis data achieved superior performance (Macro F1-score: 94.15%). SHAP analysis revealed that heavy hydrocarbon components (i-C5, n-C5) were the most discriminative features for oil-bearing zones, while also identifying complex, non-linear relationships between certain features (e.g., oil index) and different fluid types. The developed methodology provides a rapid, robust, and objective approach for reservoir fluid identification, with the complete workflow offering valuable insights for feature selection and model construction in similar petrophysical studies.