A COMPARATIVE STUDY FOR WEB SERVICES RESPONSE TIME PREDICTION

Abstract

Web-based learning communities could significantly benefit from having as feature adaptive systems. Those models take advantage of the knowledge and experiences of community affiliates and utilize it to better serve each person. Learning is naturally a practice directly associated to sociability. Conventional learning involves the development and operation of a community. Many research studies offer evidence that "strong feelings of community may not only increase persistence in courses but may also increase the commitment to group goals, cooperation among members, satisfaction with group efforts, and motivation to learn" (A. P. Rovai et al., 2002. Sometimes a community may fall into more than one classification and eventually could develop sub communities created around particular interest groups. Learning communities are classically considered communities of purpose, their function being learning. The beginning of online communities has shown to be a fairly promising idea, permitting the enhancement of both the quality of online courses and the pleasant appearance of web-based learning environments. Designing and implementing an online setting needed by a community requires more them just communication and resource sharing capabilities (Maria Rigou, 2004). Nowadays web-based learning communities, separately from offering high quality content, efficient structuring and support for the everyday jobs of all users, have radically developed and included techniques from other domains like adaptive systems and artificial intelligence. In his study we compare three recurrent adaptive models in order to show which one offers higher prediction accuracy of web services response time. Nowadays web-based learning environments must provide high quality content with lower response time in order to achieve the needs of a modern online learning community. The rest of the paper is structured as follows: the next section describe the recurrent neural networks and the three particular models we compare in this study; the third section describes the experimental system and the function of its components; the following section presents the results and the study is ended with relevant conclusions and future working plans. Recurrent neural networks have proven their predictive capability on time series. In this study, we compare the prediction accuracy of three recurrent networks on a data set that consists of real world web services response time. We also have investigated the effect of encoded inputs on the prediction accuracy of the three models.