A New Outlier Detection Algorithm Based on Kernel Density Estimation for ITS

Intelligent Transportation System (ITS) uses traffic data gathered by crowdsensing technology, which can easily get vast amounts of data from ordinary people's mobile devices, to ease congestion. However, crowdsensing also highlights the problem that the abnormal data, which we often call as outliers, may be collected for analyzing and then decrease the performances of ITS. To deal with this problem, we propose a new outlier detection algorithm based on Kernel Density Estimation (KDE) in this paper. Firstly, an optimal estimation of the traffic data's probability density function (PDF) is acquired by KDE. Then a belief function, which is determined by PDF, is built to detect the outliers in the dataset. Simulation results indicate that, compared with the traditional outlier detection algorithm in ITS, our algorithm achieves higher detection rate and lower false detection rate.