Modern internal combustion (IC) engines have complex configurations with many parameters to be tuned, and this system complication usually leads to a large amount of engine calibration experiments. In this study, the active subspace method was used to build predictive models for the gas exchange-related parameters, including volumetric efficiency, intake mass flow and pumping loss, and the power-related parameters, including engine torque and engine power. The results show that the predicted outputs fit well with the experimental data, with satisfactory coefficients of determination and average absolute errors (AAE). Further, the contributions and influence directions of the input parameters to the outputs were provided based on a sensitivity analysis, which is consistent with the existing knowledge, and therefore, verifies the reliability of the predictive model built based on the active subspace method. Finally, the relation between the training group size and the prediction performance was explored. It is shown that a reduction, up to 66%, in the training group size is still able to maintain good predictive performances of the models, indicating the substantial capability of the active subspace method to reduce the experimental efforts.