Finding a parameter that can accurately identify the order-disorder phase transition, especially for complex physical systems with high-dimensional configurational space, is a challenging task. Recent work proposes a machine learning approach to effectively tackle this challenge.