Study of recursive relative regression predictor

Abstract

In this work, we develop first the recursive relative regression estimator using independent identically real random variables. The classical regression estimators may give unreliable prediction results with the existence of aberrant values. The recursive relative regression estimator works better in this case, further it allows the estimation to be updated whenever new observations are included in the estimation at any time. We provide next its almost complete convergence by indicating its rate. In a simulation study, we employ the recursive relative estimator as a predictor and assess its performance against various nonparametric estimators to substantiate our results.