A novel approach to forecast traffic congestion using CMTF and machine learning

Abstract

Traffic congestion severely affects many cities around the world causing various problems like fuel wastage, increased stress levels, delayed deliveries and monetary losses. Therefore, it is urgent to make an accurate prediction of traffic jams to minimize these losses. But forecasting is a real challenge to obtain promising results for vibrant and ambiguous traffic flows in urban networks. This paper proposes a new traffic congestion model using pre-calculated density from node information table based on previous traffic data. In this model, we predicted traffic congestion of an intersection according to its adjacent road's node information table, where node information table contains the traffic density of all incoming lanes of an intersection (node). Besides, for this model, we consider all intersections of a city as individual nodes, and we prepare node information table for each node. Our work can be divided into two parts: (1) we perform time series analysis on previous data of a node and its adjacent nodes, and (2) then apply those calculated values to this model and make the prediction based on it. The forecasted value will always be between 0 and 1. Where 0 means no traffic congestion, close to 0 means low traffic congestion and 1 means heavy traffic or close to 1 means congested traffic lane accordingly.