Feature Selection Method Based on the Adaptive Genetic Algorithm-Kernel Partial Least Squares for High Dimensional Data

Feature selection for modeling the high dimensional data, such as the near-infrared spectrum (NIR) is very important. A novel modeling approach combined the adaptive genetic algorithm-kernel partial least squares (AGA-KPLS) is proposed. The KPLS algorithm is used to construct nonlinear models with the popular kernel based modeling technology. The AGA is used to select the optimal feature sub-set from the original high dimensional data, which also used to select the kernel parameters of the KPLS algorithm simultaneously. The experimental results based on the vibration frequency spectrum show that the proposed approach has better prediction performance than the normal GA-PLS method.