Despite tremendous research advances in detecting Alzheimer's disease (AD), traditional diagnostic tests remain expensive, time-consuming or invasive. The search for a low-cost, rapid, and minimally invasive test has marked a new era of research and technological developments toward establishing blood-based AD biomarkers. The current study has employed excitation-emission matrices (EEM) of fluorescence spectroscopy combined with machine learning to diagnose AD using blood plasma samples from 230 individuals (83 AD patients from 147 healthy controls). To evaluate the performance of the classification algorithms, we calculated the commonly used figures of merit (accuracy, sensitivity and specificity) and figures of merit that take into account the samples unbalance and the discrimination power of the models, as F2-score (F2), Matthews correlation coefficient (MCC) and test effectiveness ([Formula: see text]). The classification models achieved satisfactory results: Parallel Factor Analysis with Quadratic Discriminant Analysis (PARAFAC-QDA) with 83.33% sensitivity, 100% specificity, 86.21% F2; and Tucker3-QDA with 91.67% sensitivity, 95.45% specificity and 91.67% F2. In addition, the classifiers show high overall performance with 94.12% accuracy and 0.87 MCC. Regarding the discrimination power between healthy and AD patients, the classification algorithms showed high effectiveness with the mean scores separated by three or more standard deviations. The PARAFAC's spectral profiles and the wavelength values from both models loading profiles can be used in future research to relate this information to plasma AD biomarkers. Our results point to a rapid, low-cost and minimally invasive blood-based method for AD diagnosis.