A new method of model prediction of active substance contents in pharmaceutical tablet based on near-infrared spectroscopy (NIR), artificial neural network regression (NNR) and independent component analysis (ICA) was proposed. In its application to tablet near-infrared spectroscopy, the independent components and the mixing matrix were firstly extracted by ICA. And then, the model of the mixing matrix and active substance content matrix was built by the BP NNR. The influence of the numbers of independent components and the neurons in the hidden layer on the properties of the model was further discussed, and three classes of tablet optimal quantitative-analysis models were built respectively. This new chemometric method has been applied to the prediction of active substance contents in three classes of pharmaceutical tablet samples. The correlation coefficients between the chemically tested values and the NIR method predicted values of active substance contents to be 0.962, 0.980 and 0.979, respectively. The research result shows the feasibility of establishing the models with ICA-NNR method for tablet quantitative analysis in pharmaceutical industry.

Key words: independent component analysis, neural network regression, near infrared spectroscopy, pharmaceutical tablet