MATLAB-Based Lightweight Workload Prediction via Machine Learning Models in Distributed Systems

This paper presents a dynamic selector model for data workload prediction. A main function responsible for selecting the most accurate Machine Learning Algorithm (e.g., Linear Regression, Support Vector Regression, and Random Forest Regression) has been developed. The selection of the algorithm is based on the runtime workload of the given data application ensuring an efficient and accurate development of the application. The selector model is parallelized and deployed in Docker containers which can speed up the runtime and improve the integrated performance of the system on a large scale. Evaluated results show significant improvements in scalability, versatility, and prediction accuracy by analyzing various data sets in different distributed environments.