In order to overcome the limitations of BADA-based fuel consumption models in terms of low prediction accuracy, and to address the lack of interpretability in machine learning approaches such as back propagation Neural Network (BPNN), an aircraft fuel consumption model based on the multivariate adaptive regression splines (MARS) method was developed and analyzed. The model was applied to the B737-800 aircraft during the cruise phase to investigate the nonlinear relationships between fuel flow and relevant flight parameters. Comparative analyses were conducted with three classical regression models—random forest regression (RFR), BPNN, and multiple linear regression (MLR). The results show that the MARS model achieves superior accuracy in both training and testing datasets, satisfying the requirements for predictive performance. Moreover, the MARS model provides high interpretability. The variable importance analysis indicates that aircraft weight is the most influential factor affecting fuel consumption, followed by ambient temperature. It is concluded that the MARS-based modeling approach effectively balances prediction accuracy and model interpretability, offering a promising tool for aircraft performance analysis and fuel optimization.