IntroductionIn recent years, machine learning has become widely used in scientific contexts, producing strong predictive results. However, an increasingly clear limitation has emerged: many models perform well numerically but remai