Discriminant Analysis Method Of Radix Astragali And Preparations Based On Multi-source Spectral Fusion Technology Combined With Random Forest

Radix Astragali is a traditional Chinese herbal medicine with homology of medicine and food,which has some important pharmacological effects such as enhancing immunity,anti-oxidation,anti-aging,anti-tumor,lowering blood sugar,protecting liver and kidney,etc.It plays a vital role in the prevention and treatment of diseases.With the increase of medicinal value and market demand of Radix Astragali,the quality control and assessment of Radix Astragali are of great significance.The sample preparation of traditional analytical methods is complicated,and it is difficult to achieve real-time and rapid analysis of the quality of traditional Chinese medicines.Laser-induced breakdown spectroscopy(LIBS)and infrared spectroscopy(IR)analysis techniques have the advantages of without complex sample pretreatment,simple operation and rapid analysis,which have been effective tool for quality control and assessment of traditional Chinese medicine.Nevertheless,owing to complexity and diversity of the matrix and chemical components of Radix Astragali and preparations,the complex and high-dimensional LIBS or IR spectrum not only contains the critical information of the target,but also contains a lot of interference information,which makes it impossible to obtain satisfactory analysis results by traditional analysis methods.Artificial intelligence and chemometrics can be applied to extract valuable information from complex spectral data so as to achieve accurate analysis of the complex system of Radix Astragali and its preparations.We developed a quality evaluation method of Radix Astragali and its preparations by the combination of LIBS,IR and chemometrics,and focusing on the discriminant analysis of Radix Astragali and its preparations by spectral fusion technology combined with random forest(RF).It can provide a new idea and method for quality control and rapid evaluation of traditional Chinese medicine.The main research contents of this article are as follows:(1)Taking Radix Astragali as the research object,principal component analysis(PCA)and RF discriminant models were constructed with LIBS,IR and data fusion.Firstly,the unsupervised discriminant model on basis of PCA was applied to analyze the LIBS,IR and fusion spectra to realize the preliminary discrimination of Radix Astragali.Then,the pretreatment method of LIBS spectra was optimized on basis of cross-validation,and the RF discriminant model was constructed on basis of the optimal spectral pretreatment method.Next,the pretreatment method of IR spectra was optimized on basis of cross-validation,and the RF discriminant model was established on basis of the optimal pretreatment method and variable importance measurement(VIM).Subsequently,the RF discriminant model on basis of the low-level data fusion of LIBS and IR was established and compared with the predictive performance of the RF model on basis of LIBS and IR.Finally,the RF discriminant models were constructed with elementary and intermediate data fusion based on VIM.The best model was selected by comparing different data fusion strategies.The results show that compared with single LIBS or IR spectra,the RF discriminant models based on data fusion strategy show better discrimination accuracy.Among them,the RF model based on the low-level data fusion of VIM not only obtains better discrimination results,but also has high computational efficiency and the sensitivity,specificity and accuracy are 0.9667,0.9833 and0.9778,respectively.Therefore,a discriminant analysis method based on VIM for LIBS and IR elementary spectrum fusion combined with RF was established,which provided a new method and idea for the identification of Radix Astragali.(2)Taking Radix Astragali granules as the research object,RF discriminant models were constructed with LIBS,IR and data fusion.Firstly,the pretreatment method and variable importance threshold of the LIBS spectra of Radix Astragali granules were optimized based on the cross-validation.And the RF discriminant model was constructed based on the optimal pretreatment method and extracted variables.Then,the pretreatment method and variable importance threshold of the IR spectra of Radix Astragali granules were optimized based on the cross-validation.And the RF discriminant model was constructed based on the optimal pretreatment method and extracted variables.Finally,the RF discriminant models based on the elementary and intermediate data fusion of LIBS and IR spectra were further established based on VIM and compared with the predictive performance of the RF models based on LIBS and IR.The results show that compared with single spectrum,both data fusion methods can effectively improve the predictive performance of the model.The model based on intermediate data fusion achieves the best prediction performance and the sensitivity,specificity and accuracy are 0.9778,0.9886 and 0.9852,respectively.Therefore,a discriminant analysis method based on LIBS and IR for intermediate spectral fusion combined with RF algorithm was established,which provided a new method and idea for the identification of Astragalus granules.