A novel method utilizing state space neural network (SSNN) with adaptive filters is proposed to estimate the traffic flow parameters. The SSNN's network topology is derived from delays and stops estimation problem, so the design of SSNN reflects the relationships that exist in physical traffic systems. To improve SSNN effectiveness, the adaptive filters is proposed to train the SSNN instead of conventional approaches. Model performance was tested with raw traffic data of an intersections group at Odem. Performance of the proposed model is compared with that of SSNN and BP neural network. Results of the comparisons indicate that the proposed model predicts complex nonlinear delays and stops with satisfying effectiveness, robustness and reliability.