Predicting public transit arrival: A nonlinear approach