Abstract
In this research, a new improvement of the Nadaraya-Watson kernel non parametric regression estimator is proposed and the bandwidth of this new improvement is obtained depending on the three different statistical indicators: robust mean, median and harmonic mean of kernel function instead of using geometric and arithmetic mean, or R. Simulation study is presented, including comparisons with four others Nadaraya-Watson kernel estimators (classical methods). The proposed estimator in the case of harmonic mean is more accurate than all classical methods for all simulations based on MSE criteria.