Model Optimisation for Server Loading Forecasting with Genetic Algorithms

Abstract

Server load prediction has different approaches and applications, with the general goal of predicting future load for a period of time ahead on a given system. Depending on the specific goal, different methodologies can be defined. In this paper, we study the use of temporal factors, along with its manual or optimised application using genetic algorithms. The main steps involved are data transformations, a novel pre-processing method based on enrichment of data through the inducing of temporal factors and a genetic algorithms wrapper that optimises all the variables in our approach. The created model was tested on a short-term load forecasting problem, with the use of data from single and combined months, regarding real data from Wikipedia servers. The learning methods used for creating the different models were linear regression, neural networks, and support vector machines. A basic dataset, as well as an enriched dataset, were the core elements for the two scenarios studied. Results show that it is possible to tune the dataset features, e.g., granularity and time window to improve prediction results.