Traffic forecasting using least squares support vector machines

Abstract

Accurate and timely forecasting of traffic parameters is crucial for effective management of intelligent transportation systems. Travel time index (TTI) is a fundamental measure in transportation. In this article, a non-parametric technique called least squares support vector machines (LS-SVMs) is proposed to forecast TTI. To the best of our knowledge, it is the first time to cooperate the rising computational intelligence technique with state space approach in traffic forecasting. Five other baseline predictors are selected for comparison purposes because of their proved effectiveness. Having good generalisation ability and guaranteeing global minima, LS-SVMs perform better than the others. Experimental results demonstrate that our predictor can significantly reduce mean absolute percentage errors and variance of absolute percentage errors, especially for predicting traffic data with weak regularity. Persuasive comparisons clearly show that it provides a large improvement in stability and robustness, which reveals that it is a promising approach in traffic forecasting and time series analysis.

Spatiotemporal Patterns in Large-Scale Traffic Speed Prediction

“…Traffic prediction methods also follow the same assumption [2]- [9], [13], [16], [25]. Similar to other studies, we train the algorithm with 50 days of data and perform prediction for 10 days [3], [5], [13]. It is important to point out that this assumption may not hold true in the long term.…”

Low-Dimensional Models for Compressed Sensing and Prediction of Large-Scale Traffic Data

“…These methods offer greater flexibility due to their generic structure. Consequently, these methods are used to develop highly accurate traffic estimation and prediction models [8], [11], [12]. In all of these studies, data-driven techniques explicitly predict traffic variables at each link in the observed network.…”

Comparative Evaluation of The Traffic Flow Volatility Forecasting Models

“…Tremendous univariate prediction models have been proposed including parameter models such as time series models [1][2][3], Kalman filtering [4][5], support vector machine [6][7], and some non-parameter models such as nonparametric regressive model [8][9] and neural network model [10][11][12]. To further improve the short-term forecasting accuracy, some multivariate models was introduced to calibrate the relationships between different traffic flow variables at a traffic station or the same variable at different traffic stations.…”