The modified feature subsets ensemble applied for the mach number prediction in wind tunnel

Large-scale and high-dimensional data are the main difficulties for Mach number prediction in a wind tunnel system. To solve these issues, based on the multivariate fuzzy Taylor theorem, the Feature Subsets Ensemble (FSE) method has been proposed. However, data sampled from measurements have noise. The FSE uses the entire data set at least once, and for data with noise this property may reduce the generalization of Mach number prediction. In this paper, a Modified Feature Subsets Ensemble (MFSE) method is proposed to overcome the limitation of the FSE on noisy data by introducing the bootstrap to the feature subsets. The bootstrap has the potential to avoid noise. In addition, the generating procedure of the bootstrap replications is straightforward, simple, and quick, which keeps the complexity of the MFSE low. Experiments show that the MFSE outperforms the FSE and the Random Forest method, particularly on high-noise data sets. For all three working conditions, on both the low- and the high-noise testing sets, only the MFSE estimator meets the requirements of forecasting speed, accuracy, and generalization of Mach number prediction.