Abstract
This paper presents nonparametric generalized additive models to detect both lane-blocking and shoulder incidents for two freeways in Colorado and California, USA. The parametric generalized additive models were developed based on the examination of the partial prediction of the variables of the nonparametric generalized additive model. This paper also highlights the importance of developing incident detection models to detect both lane-blocking and shoulder incidents. The performance of the nonparametric generalized additive models were also compared to multilayer feedforward neural network based models. The model proposed outperformed the neural network based model.
