This thesis explores the potential of reducing travel time through sophisticated routing methods utilizing traffic speed forecasting models. The current solution used by Triona relies on a naive static model, which assumes that traffic always flows at the speed limit for all roads. While this approach yields reasonable results, it does not account for spatial or temporal variations in traffic conditions, leaving room for improvement. To address this, we develop and evaluate more advanced forecasting models including statistical approaches such as autoregressive integrated moving average and machine learning models such as LSTM and traffic graph convolutional LSTM. The methodology involves collecting historic traffic data from the Stockholm metropolitan area, preprocessing the data, adapting traffic speed forecasting models, implementing routing algorithms and evaluating generated routes. The routing algorithms, based on a modified Dijkstra's algorithm, solve the time-dependent shortest path and time-dependent traveling salesman problem. The evaluation shows that utilizing the LSTM model instead of the naive static model for routing reduces the difference from optimal routing from 4.37% to 1.18% for time-dependent shortest paths and from 5.40% to 1.52% for time-dependent traveling salesman problems. This thesis contributes a framework for applying and evaluating traffic speed forecasting methods to real traffic data, highlighting the potential increase in transportation efficiency through intelligent routing.