Several approaches to forecast the service quality based on its quality of service (QoS) properties are reported in the literature. However, their main disadvantage resides in their limited scalability. In fact, they elaborate a forecasting model for each quality attribute per service, which cannot scale well for large or even medium size datasets of services. Accordingly, we propose a novel scalable representative-based forecasting approach of QoS. The QoS is modeled as a multivariate time series in which the values of service attributes are evaluated at each time instant and forecasted based on three stages. First, a data aggregation function is applied to the multivariate time series data. Then, principal component analysis (PCA) is applied to the quality attributes to determine the most relevant ones. The reduced data is then clustered, so that, a representative for each cluster is computed. Finally, a forecasting model is built for each cluster representative for the sake of deriving other services’ forecasting models. A set of extensive experiments are carried out to assess the efficiency and accuracy of the proposed approach on a dataset of real services. The experimental results show that the proposed approach is up to 75% more efficient than direct forecasting approaches using time measurements while increasing the number of forecasted services and that the elaborated forecasting models enjoy insignificant forecasting errors.

中文翻译：

---

一种新的基于可扩展代表的服务质量预测方法

文献中报道了几种基于服务质量 (QoS) 属性来预测服务质量的方法。然而，它们的主要缺点在于其有限的可扩展性。事实上，他们为每项服务的每个质量属性制定了一个预测模型，对于大型甚至中等规模的服务数据集都无法很好地扩展。因此，我们提出了一种新颖的基于可扩展代表的 QoS 预测方法。QoS 被建模为一个多元时间序列，其中服务属性的值在每个时刻被评估并基于三个阶段进行预测。首先，对多元时间序列数据应用数据聚合函数。然后，将主成分分析 (PCA) 应用于质量属性以确定最相关的属性。然后对减少的数据进行聚类，以便计算每个集群的代表。最后，为每个集群代表建立一个预测模型，以便推导出其他服务的预测模型。进行了一组广泛的实验，以评估所提出的方法在实际服务数据集上的效率和准确性。实验结果表明，所提出的方法比使用时间测量的直接预测方法高 75%，同时增加了预测服务的数量，并且精心设计的预测模型的预测误差很小。进行了一组广泛的实验，以评估所提出的方法在实际服务数据集上的效率和准确性。实验结果表明，所提出的方法比使用时间测量的直接预测方法高 75%，同时增加了预测服务的数量，并且精心设计的预测模型的预测误差很小。进行了一组广泛的实验，以评估所提出的方法在实际服务数据集上的效率和准确性。实验结果表明，所提出的方法比使用时间测量的直接预测方法高 75%，同时增加了预测服务的数量，并且精心设计的预测模型的预测误差很小。