Given the global growth in the production and consumption of generic drugs and the increasing risk of counterfeit or substandard pharmaceutical products, the development of novel, rapid, and non-destructive quality control methods has become critically important. In this study, hyperspectral imaging (HSI) in the visible to near-infrared range (Vis-NIR, 400–950 nm), combined with chemometric/machine learning techniques, was employed to assess the active pharmaceutical ingredient (API) content of metformin in powder-based samples. Samples were classified into three dosage groups: standard dose (SD), low non-standard dose (LD), and high non-standard dose (HD). Hyperspectral imaging data were processed using spectral averaging, principal component analysis (PCA), partial least squares-discriminant analysis (PLS-DA), and machine learning algorithms including artificial neural network (ANN) and support vector machines (SVM). Results demonstrated that chemometric models, particularly ANN and PLS-DA, could effectively differentiate between the three sample groups with high accuracy. The combination of Vis-NIR HSI and statistical modelling proved to be a powerful tool for detecting metformin dosage levels and distinguishing standard from non-standard pharmaceutical compositions.