A Computational Evaluation of Distributed Machine Learning Algorithms

Over the past few years, availability of data throughout the internet has scaled to petabytes. Processing this data demands a strong computing mechanism rather than usual machine learning techniques. Distributed machine learning solves this problem with distributed system environment. It refers to multi-node machine learning algorithms and systems that are designed to scale to larger input data sizes. To understand the best possible utilization for each algorithm in a parallel computing environment, a detailed study of these distributed machine learning algorithms is necessary. To summarize and evaluate their performance, parameters such as Execution time, CPU usage, Memory usage rate and Accuracy have been considered for each algorithm. The experiment conducted on LightGBM, CatBoost, AdaBoost and XGBoost demonstrated that CatBoost performed the best with an accuracy of 81.31% amongst the other algorithms. This analysis will thus help beginners and business firms chose the appropriate algorithm for their specific distributed machine learning application.