Short-term traffic flow forecasting by mutual information and artificial neural networks

Short-term traffic flow forecasting is an important problem in the research area of intelligent transportation system (ITS). Recently, artificial neural networks modeling, such as MLP, have been used in various applications over nonlinear time series forecasting such as traffic control. In modeling, irrelevant inputs cause the deterioration of performance and increment of calculation cost. Therefore, to have an accurate model, some strategies are needed to choose a set of most relevant inputs. Mutual information (MI) is very effective in evaluating the nonlinear relevance of each input from the view of information theory. Feature selection (FS) method is an improved version of the MI technique. This paper presents a novel short-term traffic flow prediction model using MLP predictor and MIFS algorithm. Performance of the proposed MIFS algorithm and MLP predictor is evaluated via simulations using MATLAB subroutine. To validate the algorithm, two different types of data, namely regular and irregular (with high uncertainty) data, are used.