Application of locally weighted regression-based approach in correcting erroneous individual vehicle speed data

Abstract

Because of the quality of raw data being an essential feature in determining the reliability of traffic information, an effective detection and correction of outliers in raw field-collected traffic data has been an interest for many researchers. Global positioning systems (GPS)-based traffic surveillance systems are capable of producing individual vehicle speeds that are vital for transportation researchers and practitioners in traffic management and information strategies. This study proposes a locally weighted regression (LWR)-based filtering method for individual vehicle speed data. To fully and systematically evaluate this proposed method, a technique to generate synthetic outliers and two approaches to inject synthetic outliers are presented. Parameters that affect the smoothing performance associated with LWR are devised and applied to obtain a more robust and reliable data correction method. For a comprehensive performance evaluation of the developed LWR method, comparisons to exponential smoothing (ES) and autoregressive integrated moving average (ARIMA) methods were conducted. Because the LWR-based filtering method outperformed both the ES and ARIMA methods, this study showed its useful benefits in filtering individual vehicle speed data. Copyright (c) 2015 John Wiley & Sons, Ltd.