Comparative Analysis of Time Series Models for Traffic Flow Prediction

One of the important purposes for machine learning is to analyze the data and use that data in order to forecast data and many times especially in time series data, forecasting becomes very hard. The need of analyzing and forecasting traffic flow has always been emphasized. Government can collect data and use the data for better traffic management and avoid many traffic congestions. The data collected can then be analyzed and used for traffic volume forecasting which can save a huge amount of time lost in commute. It may also save lives by allowing emergency vehicles, such as ambulances, to get to hospitals more quickly. The problems stated above need a solution where it becomes at most necessary to forecast the traffic volume in order to prepare for the same. In this comparative analysis four different models are used for predicting traffic flow. The models include ARIMA, FbProphet, XgBoost and LSTM. These selected models have been used in order to forecast the hourly traffic volume. The dataset for traffic flow analysis is taken from UCI machine learning repository. The performance is then evaluated on the basis of several parameters such as Root Mean Square Error(RMSE), and Mean Absolute Percentage Error(MAPE). The XgBoost model performs the best as it achieves the lowest MAPE among all the considered models. The LSTM model performs better than the other models considered for the study.