The development of computationally efficient model selection strategies represents an important problem facing the analysis of Nuclear Fusion experimental data, in particular in the field of scaling laws for the extrapolation to future machines, and image processing. In this paper, a new model selection indicator, named Model Falsification Criterion (MFC), will be presented and applied to the problem of choosing the most generalizable scaling laws for the power threshold (PThresh) to access the H-mode of confinement in Tokamaks. The proposed indicator is based on the properties of the model residuals, their entropy and an implementation of the data falsification principle. The model selection ability of the proposed criterion will be demonstrated in comparison with the most widely used frequentist (Akaike Information Criterion) and bayesian (Bayesian Information Criterion) indicators.