Information systems resource capacity planning using machine learning algorithms

Abstract

The purpose of this research is to learn CPU and memory utilization data to derive more information from the data, that would help information technology organizations like telecoms mitigate some problems and offer opportunities for capacity planning and cost optimization for CPU and memory resource requirements for information technology systems. Classification problems for various data set concepts have been attracting a significant amount of research attention using machine learning algorithms. Raw data with attribute-value or instance-value has been the most frequently encountered data type which can be presented on a Euclidean space. Classification methods are used for discovery, recognition, grouping and prediction The CRISP-DM methodology was used for this research and various ML algorithms for data classification and prediction were reviewed and accessed. This study showed that supervised ML and in particular KNN was appropriate for the research problem. For more meaningful information and decision making, extra features on the data were required to determine resource utilization levels. Using the Weka software suite, the study showed that there were interesting prediction results which were obtained by learning the data to predict required memory and CPU loads given that some level of optimization was required. Prediction results for CPU and memory resources are discussed as well as those from tree type J48 and SVM methods are compared. Finally, recommendations and conclusion are given.