Decoding the synergy: unveiling gradient boosting regression model for multivariate quantitation of pioglitazone, alogliptin and glimepiride in pure and tablet dosage forms.

This study represents a comparison among the performances of four multivariate procedures: partial least square (PLS) and artificial neural networks (ANN) in addition to support vector regression (SVR) and extreme gradient boosting (XG Boost) algorithm for the determination of the anti-diabetic mixture of pioglitazone (PIO), alogliptin (ALG) and glimepiride (GLM) in pharmaceutical formulations with aid of UV spectrometry. Key wavelengths were selected using knowledge-based variable selection and various preprocessing methods (e.g., mean centering, orthogonal scatter correction, and principal component analysis) to minimize noise and improve model precision. XG Boost effectively enhanced computing speed and accuracy by focusing on specific spectral features rather than the entire spectrum, demonstrating its advantages in resolving complex, overlapping spectral data. The independent test results of different models demonstrated that XG Boost outperformed other methods. XG Boost achieved the lowest root mean squared error of prediction (RMSEP) and standard deviation (SD) values across all compounds, indicating minimal prediction error and variability. For PIO, XG Boost recorded an RMSEP of 0.100 and SD of 0.369, significantly better than PLS and ANN. For ALG, XG Boost showed near-perfect performance with an RMSEP of 0.001 and SD of 0.005, outperforming SVR and PLS, which had higher error rates. In the case of GLM, XG Boost also excelled with an RMSEP of 0.001 and SD of 0.018, demonstrating superior precision compared to the much higher errors seen in PLS and ANN. These results highlight XG Boost's exceptional ability to handle complex, overlapping spectral data, making it the most reliable and accurate model in this study.