Principal Component Selection for Neural Network Classification of Active Ingredients from Near Infrared Spectra

This paper presents results from an investigation into the capability of reflectance Near-Infrared (NIR) spectroscopy in the classification of pharmaceutical tablets based on their active ingredients. The part of the process selected for this study is the packaging step in the manufacture of pharmaceuticals because in older processing lines the product was classified after tablet coating and before blister packaging using visual automatic inspection techniques. However, for 100% conformance, a more discriminating technique is required. In this investigation, NIR spectra (with wavelengths between 400nm and 1100nm) were obtained for samples relating to 3 different types of pharmaceuticals: Quetiapine, Ibuprofen and Paracetamol. The dimensionality of the dataset was reduced using Principal Components and then different combinations of Principal Components were fed into a Neural Network that was configured to classify the tablets based on their active ingredients. The recognition rates achieved in this study were high enough to suggest that NIR spectroscopy is a viable method of ensuring 100% identification of pharmaceutical tablet type prior to packaging for the pharmaceuticals tested.