Response grid plots for model-agnostic machine learning insight

Abstract Accepted by: Aris Syntetos Machine learning (ML) models provide predictive characterization of response functions based on observed or simulated data. Insight on the nature of the approximation to the underlying response function is essential for the model output to be trusted and used by decision makers. When a response function is well-behaved, methods that identify marginal effects and interactions for important input variables can provide adequate insight on its nature. Alternatively, interpretable ML models can provide an ensemble of simple models to decompose a complex response into interpretable elements. This article introduces a model-generated but model-independent visual display of functional information that provides direct insight, not interpreted through model form, model coefficients, or numerical characterizations such as those produced by sensitivity analysis. Further, the method can be applied directly to system response data if the data are collected from a factorial (grid) design. The value of response grid plots (RGPs) is shown through several examples.