This document describes a new machine learning approach for solving fracture mechanics problems when analytical solutions are not available. The authors develop regression tree and neural network models to predict fracture toughness from microcantilever geometry and loading conditions. They find that both approaches provide accurate results, but neural networks outperform regression trees with predictions within 1.5% of finite element simulations. This demonstrates that machine learning solutions can overcome limitations of empirical approaches and change how engineering problems are solved when analytical solutions cannot be obtained.