• Nazifi Lawal Bashir
• Sunday Tarekakpo Odobai

Near-infrared, NIR, spectroscopy has found great applications in pharmaceuticals, food and petrochemicals through developing models that relate the spectral absorbance and sample property of interest. The classical approach to develop these models involves a univariate linear regression at a single selected wavelength. This research was aimed at developing data driven empirical models using several wavelengths for the prediction of active substance content in a pharmaceutical tablet from Near-Infrared spectral data. Prior to model computation, spectral data was pre-processed to remove unwanted spectral variations in the data and the traditional partial least squares, PLS, regression technique was used to develop benchmark models, with cross-validation number of components, used to evaluate the performance of the data-driven models. Artificial neural networks, ANN, is a data driven model computation method that can model complex datasets, this method was used to develop the empirical models. Pre-processing of the data showed significant effect in PLS and ANN models with improved model performance observed when data is first pre-processed before model computation. Models developed using ANN performed better than the models developed using PLS with higher correlation coefficient and lower root mean squared error of prediction. In this research a combination of PLS and ANN in computing multivariate models is proposed where the PLS predictor scores with reduced dimension are used as input for computing the ANN model. By comparison of the models developed, the combination of PLS and ANN resulted in the best model with coefficient of determination of 0.97 and root mean squared error of prediction of 0.22. This therefore illustrates the potential application of combination of PLS and ANN in modelling NIR spectral data.

Machine Learning, Supervised Learning, Near Infrared Spectroscopy, Chemometrics, Partial Least Square, Artificial Neural Network.

"DATA ANALYTICS – NEAR INFRARED SPECTRAL DATA", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.9, Issue 2, page no.d133-d144, February-2022, Available :http://www.jetir.org/papers/JETIR2202318.pdf

"DATA ANALYTICS – NEAR INFRARED SPECTRAL DATA", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.9, Issue 2, page no. ppd133-d144, February-2022, Available at : http://www.jetir.org/papers/JETIR2202318.pdf