The efficiency of data-intensive applications in distributed environments such as Cloud, Fog, and Grid is directly related to data access delay. Delays caused by queue workload and delays caused by failure can decrease data access efficiency. Data replication is a critical technique in reducing access latency. In this paper, a fuzzy-based replication algorithm is proposed, which avoids the mentioned imposed delays by considering a comprehensive set of significant parameters to improve performance. The proposed algorithm selects the appropriate replica using a hierarchical method, taking into account the transmission cost, queue delay, and failure probability. The algorithm determines the best place for replication using a fuzzy inference system considering the queue workload, number of accesses in the future, last access time, and communication capacity. It uses the Simple Exponential Smoothing method to predict future file popularity. The OptorSim simulator evaluates the proposed algorithm in different access patterns. The results show that the algorithm improves performance in terms of the number of replications, the percentage of storage filled, and the mean job execution time. The proposed algorithm has the highest efficiency in random access patterns, especially random Zipf access patterns. It also has good performance when the number of jobs and file size are increased.