Traffic prediction using machine learning: A comparison of LSTM, random forest, and gradient boosting approaches

Traffic forecasting has significance in intelligent transportation systems since it helps to effectively manage traffic congestions and planning of travel paths. This paper compares three significant contemporary trends in machine learning, including Long Short-Term Memory-based (LSTM), Random Forest (RF), and Gradient Boosting (GB) models to predict traffic volumes. Random Forest and Gradient Boosting are selection models unlike LSTM, which make use of many decision trees in making predictions. We pre-process real-life traffic data, first of all, and present it in an appropriate format that can be processed by time-series forecasting, and then we evaluate the quality of the models based on whether they are capable of predicting the traffic volumes appropriately. We also find that LSTM produces a smaller prediction error than random forests and gradient boosting and apprehends complex temporal characteristics in the data. Nevertheless, in terms of run time, random forest and gradient boosting models are almost as effective as GBRT, so it is possible to use them as substitutes in case of the high importance of running time. This piece of work was informative in ranking the various models so that the user could get to know their strengths and weaknesses and come up with the roadmap in picking the best machine learning method to use in predicting traffic.