This research aims to develop an innovative artificial intelligence approach based on neural networks to estimate future traffic conditions on freeways. The traffic data was generated from a traffic simulation model for a busy freeway in Melbourne. Then, a predictive model was developed that estimates future speed, flow and density at forecast time horizons of 15, 30, 45 and 60 minutes into the future. The results showed a prediction rate ranging from 85% (flow) through 95% (speed) to 97% for density. This provides traffic operators with a model to predict the state of congestion on a freeway with a high degree of accuracy. Current and future work is focused on enhancing the neural network performance by training the models on larger numbers of observations obtained from the simulation model under different traffic conditions. The models will then be validated on real-time data obtained from field conditions.