Mechanism Discovery and Model Identification Using Genetic Feature Extraction and Statistical Testing

One main drawback of many machine learning-based regression models is that they are difficult to interpret and explain. Mechanism-based first-principles models, on the other hand, can be interpreted and hence preferable. However, as they are often quite challenging to develop, the appeal of machine learning-based black-box models is natural. Here, we report a genetic algorithm-based machine learning system that automatically discovers mechanistic models from data using limited human guidance. The advantage of this approach is that it yields simple, interpretable, features and can be used to identify model forms and fundamental mechanisms that are often seen in chemical engineering. We demonstrate our system on several case studies in reaction kinetics and transport phenomena, and discuss its strengths and limitations.