Investigating the effect of global metrics in travel time forecasting

The effect of traffic in routing, either for individuals or fleets, becomes more and more noticeable as the social, economical, and the ecological effects that it has, seem to be crucial. Forecasting travel times is an interesting, yet challenging problem, which if taken into careful consideration, could have a positive impact on the effectiveness of Intelligent Transportation Systems. Upon analyzing the problem and describing its variances, this paper compares different methodologies on traffic prediction, along with analyzing the effect of metrics, such as Principal Component Analysis and Cross Correlation, when interpreting traffic data. We evaluate known literature methods along with a new prototype algorithmic variation of STARIMA, based on the use of global Coefficient of Determination, against two diverse datasets. The benchmarking results, which are promising, are discussed with respect to the distinct characteristics of the two datasets.