Coffee is the most consumed beverage and the second most valuable traded commodity in the world. In this current study, a pocket-sized spectrometer and multivariate analysis were used for rapid authentication of coffee varieties (Arabica and Robusta) in three states to check mislabelling (food fraud). Two main coffee varieties were collected from different locations in Africa. The samples were scanned in the 740-1070 nm wavelength and the spectral data were pre-treated with several methods: mean centering (MC), multiplicative scatter correction (MSC), first derivative (FD), second derivative (SD) and standard normal variate (SNV) independently while partial least squares discriminate analysis (PLS-DA), K-nearest neighbour (KNN) and support vector machine (SVM) were used to comparatively build the prediction models for coffee beans (raw, roasted and powdered). The performances of the models were evaluated by using accuracy and efficiency. Among the classification methods developed, the best results were obtained for the following: raw coffee bean SD-SVM had an accuracy of 0.92 and efficiency of 0.82. For roasted coffee beans, SD-KNN had an accuracy of 0.92 and efficiency of 0.87, while for roasted powdered coffee, FD-KNN showed an accuracy of 0.97 and efficiency of 0.97. These finding reveals that for a more accurate differentiation of coffee beans, the roasted powder offers the best results. The obtained results showed that a pocket-sized spectrometer coupled with chemometrics could be employed to provide accurate and rapid authentication of different categories of coffee bean varieties.