Lasso Regularization Paths for NARMAX Models via Coordinate Descent

TL;DR

This paper introduces a coordinate descent algorithm for estimating NARMAX models with L1 regularization, effectively handling error regressors and producing sparse, interpretable models efficiently.

Contribution

It extends Lasso estimation to NARMAX models by including error regressors, enabling sparse and interpretable nonlinear models with an efficient algorithm.

Findings

The algorithm computes the entire regularization path efficiently.

It handles error regressors without reducing computational speed.

The method successfully identifies key regressors in linear and polynomial models.

Abstract

We propose a new algorithm for estimating NARMAX models with $L_1$ regularization for models represented as a linear combination of basis functions. Due to the $L_1$-norm penalty the Lasso estimation tends to produce some coefficients that are exactly zero and hence gives interpretable models. The novelty of the contribution is the inclusion of error regressors in the Lasso estimation (which yields a nonlinear regression problem). The proposed algorithm uses cyclical coordinate descent to compute the parameters of the NARMAX models for the entire regularization path. It deals with the error terms by updating the regressor matrix along with the parameter vector. In comparative timings we find that the modification does not reduce the computational efficiency of the original algorithm and can provide the most important regressors in very few inexpensive iterations. The method is illustrated for linear and polynomial models by means of two examples.