Nowadays, Big Data has attracted considerable attention from individuals as well as companies because of its real benefits, such as better decision-making capabilities about useful information. But this useful information can’t be extracted using classical programming and low-power processing. In this context, specific technologies (hard and soft) have to be taken into consideration. Therefore, this paper aims to filter a specific benchmark Big dataset called UNSW-NB15 using multithreading in High-Performance Computing (HPC). The obtained filtered dataset can be trained and tested by many applied approaches in critical domains with the purpose to improve the performance metrics. The experimental results show that our study improves a good performance by applying multithreading in comparison to monothreading with and without HPC. A supply comparative study is done to compare the filtered dataset and the original one. Our filtered dataset seems to be the best for training some well-known machine learning (ML) techniques for real-time attack detection so-that they can achieve 100% with the lowest time taken to build the model (0.05 s in the case of NB).