In a previous couple of decade, we've seen associate degree info explosion era and for that giant quantity of distributed knowledge being kept. The number of applications supported Apache Hadoop is increasing, the explanation behind this is often lustiness and dynamic options of this technique. HDFS provides high availability and reliableness. Characteristics of parallel operations on the application layer, Access rate is completely different for every file in HDFS. HDFS uses cloud storage to implement the functionality. By considering the assorted drawbacks of replication system of HDFS, this paper implements an approach to flexibly replicate the data files. Predictive analysis is used for setting the replication factor. The frequent access files are often replicated consistently with their access potential. Therefore, our approach concurrently improves the performance of HDFS and maintains high accessibility of data files.