Reversible data hiding (RDH) algorithms are concerned with the recovery of the original cover image upon the extraction of the secret data. In some RDH algorithms, embedding and extraction are performed through modifying the histogram of prediction errors that are computed using certain predictors. However, the embedding capacity and image quality in these algorithms are dependent on the characteristics of a single predictor. In this paper, we propose the idea of employing multiple predictors in order to increase the embedding capacity. The idea of multiple predictors is integrated in the modification of prediction errors algorithm without adding any overhead regarding the predictor that is used for embedding at each pixel in the image. The selection of the accurate predictor depends on the polarity of the prediction errors from all predictors. Experimental results proved the efficiency of the proposed algorithm in terms of increasing the embedding capacity and producing competitive quantitative and qualitative quality of the stego image when compared with recent algorithms.