In today’s world, data analysis has playedplays an important role in the performance of the systems; because it helps to improve the system’stheir function. One of the most important data mining algorithms is anomaly detection. Anomaly detection, which is a behavior in detecting system abnormality that helps finding system problems and troubleshooting. Intrusion detection and fraud detection in credit cards are some examples of anomaly detection in real worldlife. According to the increasing data volumes of the datasets that createscreate the big data, traditional data mining approaches do not have efficient results. So far, different platforms, frameworks and algorithms for big data mining have been presented. For instance, Hadoop and Spark are some of the most used frameworks in this field. Support Vector Machine (SVM) is one of the most popular approaches in anomaly detection which, according to its distributed and parallel extensions, it isis widely used in big data mining. In this research, mutual information is used for feature selection. In addition, the kernel function of the one-class support vector machine has been improved; thus, the performance of the anomaly detection has also been improved. This approach is implemented using spark. The NSL-KDD dataset is used, and an accuracy of more than 80 percent is achieved. Compared to the other similar approaches in anomaly detection, the results are improved too.