Solving the minimal attribute reduction (MAR) in rough set theory (RST) is an NP-hard combinatorial optimization problem. Ant colony optimization algorithm (ACO) is a globally heuristic optimization algorithm in evolutionary algorithms, so combining RST with ACO is an effective and feasible way to solve attribute reduction. The ACO algorithm often fall into local optimal solution, and the convergence speed is slow. This study first uses an improved information gain rate as heuristic information, and then deduction test is performed on each selected attribute and the optimal reduction set of each generation. And the mechanism of calculating probability in advance is proposed to avoid repeated calculation of information on the same path in the search process for each ant. But the algorithm can only handle small-scale data sets. The ACO algorithm has good parallelism and the equivalent classes in rough set theory can be calculated by cloud computing. This study proposes a parallel attribute reduction algorithm based on ACO and cloud computing to solve massive data sets and further investigate a multi-objective parallel solution scheme, which can simultaneously calculate the importance of the remaining attributes relative to the current attribute or reduction set. Experiments show that the proposed algorithm can obtain the MAR in the case of processing big data, and complexity of time on calculating the importance of attribute decreases from O(n²) to O(|n|).