CUDA-Driven Acceleration of Random Forest : A Solution for Scalable Machine Learning

CUDA-Driven Acceleration of Random Forest: A Solution for Scalable Machine Learning presents a comparative analysis of CPU-based serial Random Forest execution and GPU-accelerated parallel implementation using NVIDIA CUDA and RAPIDS libraries. The research evaluates performance across large-scale datasets by comparing Scikit-learn-based CPU execution with cuML and cuDF GPU execution. The study focuses on execution time, scalability, prediction latency, and computational efficiency. Experimental results demonstrate that CUDA-based GPU acceleration significantly reduces Random Forest training time while maintaining model accuracy. The research further examines the trade-offs between GPU-based training and CPU-based inference, particularly in latency-sensitive and edge computing environments. The implementation utilizes Python, CUDA Toolkit 12.8, cuML, cuDF, NumPy, and Scikit-learn within a reproducible Linux-based environment. Future work includes hybrid CPU-GPU architectures using CUDA and OpenMPI for scalable distributed machine learning systems.