Reversible data hiding (RDH) in encrypted images is an emerging research domain in the field of data security. Since the embedding rate from the existing RDH schemes in encrypted images is low, those schemes are not well suited for various real-life applications such as medical image transmission and cloud computing. To resolve this issue, in this manuscript we introduce an RDH scheme capable of providing high embedding without compromising the other efficiency parameters such as bit error rate and image recoverability. The proposed scheme uses a block-wise data hiding process in which a block of size B×B will be considered from the encrypted image at a time, and that will be scrambled through a sequence of Arnold transform. The bit sequence that the data hider wants to hide in a block will decide the number of Arnold transforming operations on the block. The same process will be continued for all the blocks in the image to get a final encrypted image with a hidden message. At the receiver side, the extraction of the hidden message and the image recovery is carried out with a trained support vector machine (SVM) model. The SVM model is capable to predict a given image block into any one of the two classes: encrypted block or natural block. The experimental study of the proposed scheme is carried out in the USC-SIPI image dataset and the results show that the new scheme surpasses the recent well-known RDH schemes.