Integration of time series models with soft clustering to enhance network traffic forecasting

The network traffic forecasting is of significant interest in many domains such as bandwidth allocation, congestion control and network management. Hence, forecasting of network traffic has received attention from the computer networks field for achieving guaranteed Quality of Service (QoS) in network. In this paper, we propose a forecasting model that combines conventional time series models with clustering approaches. The conventional linear and non linear time series models namely Weighted Exponential Smoothing (WES), Holt-Trend Exponential Smoothing (HTES), AutoRegressive Moving Average (ARMA), Hybrid model (Wavelet with WES) and AutoRegrssive Neural Network (NARNET) models are applied for forecasting network traffic. Our novelty is application of soft clustering for enhancing the existing time series models that are used to forecast network traffic. Clustering can model network traffic data and its characteristics. We derived a methodology to appropriately use cluster centriods to enhance the results obtained by conventional approach. We experimented with different soft clustering techniques such as Fuzzy C-Means (FCM) and Rough K-Means (RKM) clustering to verify the improvement in forecasting. The results of our integrated model are validated using Mean Square Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) performance measures. The results show that the integrated model enhances the results obtained using conventional time series forecasting models.