The altered glucose metabolism characterising cancer cells determines an increased amount of methylglyoxal in their secretome. Previous studies have demonstrated that the methylglyoxal, in turn, modifies the protonation state (PS) of soluble proteins contained in the secretomes of cultivated circulating tumour cells (CTCs). In this study, we describe a method to assess the content of methylglyoxal adducts (MAs) in the secretome by near-infrared (NIR) portable handheld spectroscopy and the extreme learning machine (ELM) algorithm. By measuring the vibration absorption functional groups containing hydrogen, such as C-H, O-H and N-H, NIR generates specific spectra. These spectra reflect alterations of the energy frequency of a sample bringing information about its MAs concentration levels. The algorithm deciphers the information encoded in the spectra and yields a quantitative estimate of the concentration of MAs in the sample. This procedure was used for the comparative analysis of different biological fluids extracted from patients suspected of having cancer (secretome, plasma, serum, interstitial fluid and whole blood) measured directly on the solute left on a surface upon a sample-drop cast and evaporation, without any sample pretreatment. Qualitative and quantitative regression models were built and tested to characterise the different levels of MAs by ELM. The final model we selected was able to automatically segregate tumour from non-tumour patients. The method is simple, rapid and repeatable; moreover, it can be integrated in portable electronic devices for point-of-care and remote testing of patients.