Tracking Time Evolving Data Streams for Short-Term Traffic Forecasting

摘要: Data streams have arisen as a relevant topic during the last few years an efficient method for extracting knowledge from big data. In robust layered ensemble model (RLEM) proposed in this paper short-term traffic flow forecasting, incoming data of all connected road links are organized chunks corresponding to optimal time lag. The RLEM is composed two layers. first layer, we cluster by using Graded Possibilistic c-Means method. second layer made up forecasters, each them trained forecasting on belonging specific cluster. operational phase, new chunk presented input RLEM, its memberships clusters evaluated, and if it not recognized outlier, outputs forecasters combined ensemble, obtaining way horizon. evaluated synthetic set, simulator real-world sets. gives accurate rates with outlier detection shows good adaptation non-stationary regimes. Given characteristics detection, accuracy, robustness, can be fruitfully integrated management systems.