タイトル | On the Hyperparameter Estimation of Time Varying Poisson Model for Bayesian WWW Traffic Forecasting |
---|---|
著者 | Daiki Koizumi 、Toshiyasu Matsushima 、Shigeichi Hirasawa |
年度 | 2009 |
形式 | 国際学会 |
分野 | その他 |
掲載雑誌名 | 2nd International Workshop of the European Research Consortium on Infomatics and Mathematics (ERCIM) Working Group on Computing & Statistics |
掲載号・ページ | p.22, Limassol, Cyprus |
掲載年 | 2009 |
掲載月 | 10 |
アブスト (日本語) |
学会名:2nd International Workshop of the ERCIM Working Group on Computing & Statistics 日程:29-31 October 2009 場所: Grand Resort Hotel, Limassol, Cyprus |
アブスト (英語) |
The aim is to contribute to traffic forecasting problems using a Bayesian approach for the time varying Poisson model. This model can be regarded as an application of the Simple Power Steady Model(S.P.S.M.). In this model, time variation of the parameter is formulated by a transformation function and its degree is caught by a real valued hyperparameter k (0<k<1). However, in S.P.S.M. it has not yet been proposed any definite parameter transformation function nor methods for estimation of the hyperparameter k. These two points are considered. Especially for the latter point, it has been empirically observed that World Wide Web (WWW) traffic forecasting performance strongly depends on the accuracy of the estimate of the hyperparameter k. Here at least takes two approaches for estimating the hyperparameter k, i.e. the maximum likelihood and the quasi Bayes estimations, and their effect on the traffic forecasting are discussed. According to the obtained results, the quasi Bayesian estimate gives satisfactory traffic forecasting under sufficiently large number of subintervals for integration with relatively low complexity. |
備考 (日本語) |
|
備考 (英語) |
|
論文原稿 | |
発表資料 |