Short-Term Prediction of Level of Service in Highways Based on Bluetooth Identification

A precise knowledge about future traffic will eventually open a new era in management. Research has focused on the still unresolved problem of predicting travel time (TT). However, practitioners favor level service (LOS) as meaningful metric that avoids continuous fluctuations and link-specificity TT. Evolving from TT to LOS opens research line field, moving underlying mathematical regression xmlns:xlink="http://www.w3.org/1999/xlink">classification . This study proposes short-term classifier fulfill this requirement. Given conditions are mostly free-flow throughout day, classes unbalanced. Therefore, we based our predictor Random Undersampling Boost algorithm (RUSBoost), especially suited overcome issue. We trained validated with 12 months arrival data, captured by Bluetooth network 6 links, real operation SE-30 highway (Seville, Spain). achieved an average recall 82.8% for prediction horizons up 15 minutes, reaching 92.5% congestion. reached performance exploiting two facts empirically demonstrated: (i) information every link (even those opposite direction) contributes increase accuracy prediction; (ii) presents different behavior depending day week, which used segment data construct specific classifiers. These promising results show potential proposed predictor, providing perspective into forecast subsequent management yields what demand.