Parallel Implementation of Decision Tree for Online Fraud Detection using OpenMP, MPI and CUDA

Online fraud detection is critical for ensuring the security of digital transactions. However, the growing volume of data and the need for real-time processing pose significant computational challenges. Traditional Decision Tree algorithms, commonly used for classification tasks, struggle to handle large datasets in real-time scenarios due to their computational complexity. This research addresses these limitations by parallelizing the Decision Tree algorithm to enhance its efficiency for online fraud detection. We leverage three parallel programming models—OpenMP for multi-core CPU processing, MPI for distributed systems, and CUDA for GPU acceleration to optimize the construction and evaluation phases of the Decision Tree. By distributing the computational workload across these architectures, we aim to reduce execution times while maintaining high classification accuracy significantly. A thorough performance evaluation compares the parallel implementations to traditional serial approaches, focusing on execution speed, scalability, and detection accuracy. The results reveal substantial improvements in processing speed, enabling real-time fraud detection without sacrificing accuracy. These findings provide valuable insights into developing faster, scalable, and more reliable fraud detection systems, contributing to enhanced security for financial transactions in an increasingly digital world.