Modern applications can generate a large amount of data from different sources with high velocity, a combination that is difficult to store and process via traditional tools. Hadoop is one framework that is used for the parallel processing of a large amount of data in a distributed environment, however, various challenges can lead to poor performance. Two particular issues that can limit performance are the high access time for I/O operations and the recomputation of intermediate data. The combination of these two issues can result in resource wastage. In recent years, there have been attempts to overcome these problems by using caching mechanisms. Due to cache space limitations, it is crucial to use this space efficiently and avoid cache pollution (the cache contains data that is not used in the future). We propose Hadoop-oriented SVM-LRU (HSVM- LRU) to improve Hadoop performance. For this purpose, we use an intelligent cache replacement algorithm, SVM-LRU, that combines the well-known LRU mechanism with a machine learning algorithm, SVM, to classify cached data into two groups based on their future usage. Experimental results show a significant decrease in execution time as a result of an increased cache hit ratio, leading to a positive impact on Hadoop performance.