Non-Parametric System Identification: A Comparison of MARS and Neural Networks

Feedforward artificial neural networks and multivariate adaptive regression splines (MARS) are compared in terms of their accuracy in learning different types of functions and their speed. The two methods are compared on test problems that have been used to demonstrate their efficacy. Both methods can be classified as nonlinear, non-parametric function estimation techniques, and both show great promise for fitting general nonlinear multivariate functions. We find that MARS is often more accurate and always much faster than neural networks, and develops easy-to-interpret low order models, as it heavily penalizes model complexity. However, unlike neural networks, it can also experience robustness problems with outlier responses.