Submitted to Simposio Internacional de EstadĂstica 2021, BIG DATA Y ANALĂTICA DE DATOS.
This work presents a robust version of the Nadaraya-Watson estimator, based on a weight function that penalizes the outlyingness of multivariate observations using a Mahalanobis depth measure. We compare the performance of the proposed robust version of the Nadaraya-Watson estimator using different well-known robust location and scatter estimators of the literature. The empirical results, in linear and non-linear regression, show that the proposed method is resistant to outliers and outperforms the common Nadaraya-Watson.