Verifying predictive services'quality with Mercury

Abstract

Due to the success of service technology, there are lots of services nowadays that make predictions about the future in domains such as weather forecast, stock market and bookmakers. The value delivered by these predictive services relies on the quality of their predictions. This paper presents Mercury, a tool that measures predictive service quality in the domain of weather forecast, and automates the context-dependent selection of the most accurate predictive service to satisfy a customer query. To do so, candidate predictive services are monitored so that their predictions can be eventually compared with real observations obtained from some trusted source. Mercury is a proof-of-concept to show that the selection of predictive services can be driven by the quality of their predictions. Its service-oriented architecture (SOA) aims to support the easy adaptation to other prediction domains and makes feasible its integration in self-adaptive SOA systems, as well as its direct use by end-users as a classical web application. Thoughout the paper, we show how Mercury was built.