Recently, the growing uncertainty in the power system has created an increasing need for security-constrained unit commitment (SCUC) solved on a rolling perspective (i.e., RSCUC). This puts a high requirement on the decision efficiency of the solution algorithm. To overcome the challenge, this paper proposes a POML-MCVS algorithm, which integrates a probability-oriented machine learning (POML) technique with Monte-Carlo value search (MCVS) algorithm to improve overall performance. First, POML is employed to identify the decision probabilities of units through leveraging empirical knowledge, and determine the on/off states of units with high decision probabilities. This facilitates improving the decision efficiency while maintaining the decision accuracy as much as possible. Then, MCVS is employed to optimize the rest variables and make the rolling decisions. As a modified version of Monte Carlo tree search, MCVS is able to detect the promising decisions, and evaluate their values efficiently through constructing the search tree asymmetrically, which further improves the decision efficiency. Numerical simulations conducted on a modified IEEE 39-bus system and a real 2778-bus system demonstrate the effectiveness of the proposed algorithm.