Proposing a reliable feature selection approach is the primary stone for endorsing the prediction performance; therefore, this paper proposes an enhanced optimization technique to identify the features of diverse datasets. The proposed approach employs the memory dependency of the fractional calculus operators to boost the agents’ motion during the diversification and the intensification of the hunter-prey optimization algorithm (HPO). Moreover, a chaotic sine map is applied to balance the transition between the exploitation and exploration cores, aiming to improve further convergence. As a result, a novel variant named fractional adaptive HPO (F-AHPO) is proposed as a modified feature selection (FS) model. The proposed algorithm’s applicability and performance are validated by using various UCI datasets. For the purpose of evaluating the proposed approach performance, its results are assessed and demonstrated versus a set of state-of-art techniques representing the literature. The results show the high ability of F-AHPO to enhance classification accuracy by removing irrelevant features. F-AHPO provides better results than other methods based on several performance measures.