The load variation during three-point bending (TPB) tests on prismatic Nestos (Greece) marble specimens instrumented by piezoelectric sensors is predicted using acoustic emission (AE) signals. The slope of the cumulative amplitude vs the predicted load curve is potentially useful for determining the forthcoming specimen failure as well as the indirect tensile strength of the material. The optimum artificial neural networks (ANN) model was selected based on a comparison of different machine learning techniques with respect to the root mean square error (RMSE) and the coefficient of determination (CoD). The top three best-performing techniques were decision trees, random forests and artificial neural networks. Results show that decision trees and random forests have a coefficient of determination of 98.8% and 99.2%, respectively. The artificial neural network has an accuracy of 99.6% with a root mean square error of 0.022. The comparison of results with experimental data shows that ANNs can potentially be utilized to predict rock behavior and/or establish a failure index.