Improving Urban Traffic Speed Prediction Using Data Source Fusion and Deep Learning

Abstract

Traffic parameter forecasting is critical to effective traffic management but is a challenging task due to the stochasticity of traffic flow characteristics, especially in urban road networks. Traffic networks can be affected by external factors, such as weather, events, accidents, and road construction networks. The impact of these factors can affect traffic flow parameters by influencing travel time, density, and operating speed. Although deep neural networks (DNNs) have recently shown promising signs in traffic prediction using big data, there still exists the issue of maximizing the use of the model capabilities by using big data sources. This paper proposes an improved urban traffic speed prediction approach involving input-level data fusion and deep learning. Motivated by deep learning prediction methods, we propose a Long Short-Term Memory Neural Network (LSTM-NN) for traffic speed prediction that combines traffic and weather datasets on an urban road network in Greater Manchester, United Kingdom. The experimental results substantiate the value of the approach when compared to the use of traffic-only data sources for traffic speed prediction.

A deep-learning model for urban traffic flow prediction with traffic events mined from twitter

“…We begin with a description of autoencoders, including their underlying logic and functions, before concluding with the proposed deep bi-directional LSTM prediction model. The proposed model is an enhanced version of the one presented in [11], where we employed stacked autoencoders for model training in addition to tweet messages being included as additional inputs to the model training dataset. The main advantage of an autoencoder is to learn a compressed representation (encoding) of a set of input data vectors.…”

“…We, therefore, present an approach towards urban traffic flow prediction that utilises information from tweet feeds that can contain information about non-recurring traffic events, in addition to traffic and weather-related datasets. The study in [11] displayed improved model predictive performance when weather-related (rainfall and temperature) datasets were integrated for urban traffic speed prediction. The approach presented in this current paper exemplifies an augmented and enhanced version of the one presented in [11], which is aimed at improving predictive accuracy.…”

“…The study in [11] displayed improved model predictive performance when weather-related (rainfall and temperature) datasets were integrated for urban traffic speed prediction. The approach presented in this current paper exemplifies an augmented and enhanced version of the one presented in [11], which is aimed at improving predictive accuracy. The focus on tweets as ancillary representations denoting traffic status stems from the tendency of Twitter users to react more quickly to events when compared to other social media platforms [31].…”

A Deep Learning Framework for Univariate Time Series Prediction Using Convolutional LSTM Stacked Autoencoders

“…For this study, we used the prediction setup described in [9]. In particular, the procedure involves the application of an overlapping sliding window approach for reconstructing the input time series from a univariate time series to a supervised learning format.…”

“…A number of techniques and algorithms have been proposed for time series prediction, such as linear models, which include (but are not limited to) Auto-Regressive Integrated Moving Average (ARIMA) and its variants [1], support vector machines [2], statistical analysis [3] and, more recently, deep non-linear neural network algorithms like Recurrent Neural Networks (RNN) [4], LSTMs [5] and CNNs [6], which have been applied in many areas such as in financial prediction [7], [8], traffic prediction [9]- [11], machine fault prognosis/diagnosis [12] and anomaly detection [13], [14]. Although ARIMA and ARIMA-based model variants such as Seasonal ARIMA (SARIMA) [1], Vector ARIMA (ARIMAX) [15] have shown promising signs when applied towards univariate and multivariate time series prediction, they however show vulnerabilities when applied to non-linear, sequential, or time series data, such as traffic and stock prediction [9]. In recent literature, the trend is inclined towards the use of deep learning algorithms for short to medium-term time series forecasting.…”