Short-term traffic speed forecasting based on data recorded at irregular intervals

As demand for proactive real-time transportation management systems has grown, major developments have been seen in short-time traffic forecasting methods. Recent studies have introduced time series theory, neural networks, genetic algorithms, etc., to short-time traffic forecasting to make forecasts more reliable, efficient and accurate. However, most of these methods can only deal with data recorded at regular time intervals, thereby restricting the range of data collection tools to loop detectors or other equipment that generate regular data. The study reported here represents an attempt to expand on several existing time series forecasting methods to accommodate data recorded at irregular time intervals, thus ensuring these methods can be used to obtain predicted traffic speeds through intermittent data sources such as the GPS. The study tested several methods using the GPS data from 480 Hong Kong taxis. The results show that the best performance is obtained using a neural network model with acceleration information predicted by ARIMA model.