Immune-based abnormal sample identification for spectral quantitative analysis

The abnormal samples would affect the accuracy of the spectral prediction model. This paper proposed an immune-based abnormal sample identification method for spectral quantitative analysis. The proposed method can realizes the multi-population and the multi-thread process based on the combination of the parallel computing and immune clonal selection. With evolving, the antibody populations perform the clonal selection and determine the best antibody to achieve the optimization of main population. Based on the final results, the proposed method identifies the abnormal samples, and then the partial least squares(PLS) is adopted to build prediction model based on the spectral dataset with removed the abnormal samples. Compared with the traditional immune clonal selection algorithm with PLS method and genetic algorithm with PLS method, a real near infrared spectral dataset was used in experiments, and the proposed method made the prediction residual error sum of square values of all components decreased. The experimental results verified that the proposed method can realize the identification of the abnormal samples correctly and had higher prediction precision.