CRAMP: Categorizing Classifiers and Regressors for Scalable Parallelism on Distributed and Multicore Systems

We present CRAMP (Classifiers and Regressors Adaptability Mapping for Parallelism), a unified taxonomy and benchmarking framework that categorizes machine learning models based on their adaptability to four parallelization strategies: MapReduce, multicore processing, multithreading, and gradient-based optimization. Unlike prior work, CRAMP introduces a novel parallelization-aware taxonomy linking algorithmic structure to execution behavior - an underexplored area in the literature. Experiments on the MNIST classification dataset and a real-world Alberta climate regression dataset, conducted on Google Colab, show that MapReduce excels for ensemble classifiers, multicore and multithreading boost CPU-bound models like SVMs and random forests, and gradient-based parallelism suits iterative models such as neural networks. Across both tasks, parallelization generally preserves or improves predictive performance, offering practical guidance for scalable model deployment across diverse computing architectures.