Evaluation of the employment of machine learning approaches and strategies for service recommendation

Abstract

Service functionality can be provided by more than one service consumer. In order to choose the service with the highest benefit, a selection based on previously measured experiences by other consumers is beneficial. In this paper, we present the results of our evaluation of two machine learning approaches in combination with several learning strategies to predict the best service within this selection problem. The first approach focuses on the prediction of the best-performing service, while the second approach focuses on the prediction of service performances which can then be used for the determination of the best-performing service. We assessed both approaches w. r. t. the overall optimization achievement relative to the worst- and the best-performing service. Our evaluation is based on data measured on real Web services as well as on simulated data. The latter is needed for a more profound analysis of the strengths and weaknesses of each approach and learning strategy when it gets harder to distinguish the performance profile of the service candidates. The simulated data focuses on different aspects of a service performance profile. For the real-world measurement data, 97% overall optimization achievement and over 82% best service selection could be achieved within the evaluation.