An automated forecasting method for workloads on web-based systems : Employing an adaptive method using splines to forecast seasonal time series with outliers

Abstract: This thesis introduces an automated forecasting method of time series for popular websites type of workload, such as the Wikimedia workload. These type of workloads are characterized by slowly, pronounced, changing seasonal pattern with occasional missing values and extreme outliers. The prediction method captures the seasonal pattern by cubic splines and predicts the residual by an autoregressive model. The parameters of the model are estimated from the recent observed values, outliers excluded, since detection and prediction of outliers are handled separately. The method is evaluated on the Wikimedia data, where the data is hourly based. The Wikimedia data consists of the amount of requests to the specic homepages owned by Wikimedia and the data sent from the homepages to the users internet browsers.