Dong Hu, Shuai Lyu, Shih Yu Chang, Limei Peng, P. Ho
{"title":"Dynamic SIoT Network Status Prediction","authors":"Dong Hu, Shuai Lyu, Shih Yu Chang, Limei Peng, P. Ho","doi":"10.33969/j-nana.2022.020203","DOIUrl":null,"url":null,"abstract":"Prediction of social IoT (SIoT) data traffic is helpful in characterizing the complicated relationships for such as device-to-device, user-to-user, and user-to-device. One of the most popular traffic prediction methods in noisy environments is the Kalman filter (KF), which is extremely simple and general. Nevertheless, KF requires a dynamic traffic and measurement model as a priori, which introduces extra overhead and is often difficult to obtain in reality. In comparison, deep learning models with a Recurrent Neural Network (RNN) structure have been used extensively in modeling dynamic models evolving over time. On the other hand, the Content Adaptive Recurrent Unit (CARU) is an improvement of RNN that uses fewer parameters than the LSTM and GRU and thus is more promising in predicting the SIoT data traffic. This paper proposes the CARU-based extended Kalman filter (CARU-EKF) model, which is a new deep learning cell that utilizes CARU to predict extended Kalman filter (EKF) system parameters. Note that EKF is proper to predict nonlinear SIoT traffic in noisy environments. The proposed CARU-EKF can improve the performance of time-series data forecasting for nonlinear SIoT data traffic. Numerical experiments are conducted to evaluate the SIoT traffic prediction performance of the proposed CARU-EKF approach over two real datasets, i.e., IoT device traffic and wikipedia webpage visiting traffic. The proposed method shows better performance than existing prediction methods in terms of metrics of Mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE) and determination coefficient (R2).","PeriodicalId":384373,"journal":{"name":"Journal of Networking and Network Applications","volume":"24 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Networking and Network Applications","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.33969/j-nana.2022.020203","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Prediction of social IoT (SIoT) data traffic is helpful in characterizing the complicated relationships for such as device-to-device, user-to-user, and user-to-device. One of the most popular traffic prediction methods in noisy environments is the Kalman filter (KF), which is extremely simple and general. Nevertheless, KF requires a dynamic traffic and measurement model as a priori, which introduces extra overhead and is often difficult to obtain in reality. In comparison, deep learning models with a Recurrent Neural Network (RNN) structure have been used extensively in modeling dynamic models evolving over time. On the other hand, the Content Adaptive Recurrent Unit (CARU) is an improvement of RNN that uses fewer parameters than the LSTM and GRU and thus is more promising in predicting the SIoT data traffic. This paper proposes the CARU-based extended Kalman filter (CARU-EKF) model, which is a new deep learning cell that utilizes CARU to predict extended Kalman filter (EKF) system parameters. Note that EKF is proper to predict nonlinear SIoT traffic in noisy environments. The proposed CARU-EKF can improve the performance of time-series data forecasting for nonlinear SIoT data traffic. Numerical experiments are conducted to evaluate the SIoT traffic prediction performance of the proposed CARU-EKF approach over two real datasets, i.e., IoT device traffic and wikipedia webpage visiting traffic. The proposed method shows better performance than existing prediction methods in terms of metrics of Mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE) and determination coefficient (R2).