Please use this identifier to cite or link to this item:
|Title:||Analysis and prediction of traffic congestion and incident duration in a road network||Authors:||Kalyanaraman Manikandan Jananni||Keywords:||DRNTU::Engineering::Electrical and electronic engineering||Issue Date:||2018||Abstract:||It has been observed that non-recurrent incidents such as accident, vehicle breakdown, heavy rainfall etc., lead to traffic congestion and the duration for which these incidents persist has a major impact on the roadway traffic. Thus, such traffic incidents will affect the day to day life of a commuter directly or indirectly. Also as there is always an increase in usage of the roadway network due to the ever-growing population there arises the need for an intelligent and proper traffic management system that would provide better traffic congestion control. The main objectives of Intelligent Transportation System are to predict the spread of congestion and also predict the duration to which an incident will have an impact on the roadway network so that a proper traffic management control can be implemented. Utilizing the San Francisco traffic data set, this project focuses mainly on two aspects of such predictions. Firstly, forming the classes based on queue length (number of upstream links affected) and then we concentrate on predicting the class to which traffic incidents belong using classification models like Classification and Regression Tree (CART), Support Vector Machines (SVM), Tree Bagger, K-NN classifier. Then the best classifier is determined by comparing their classification accuracies. Secondly, the thesis aims at predicting the incident duration using a prediction model that uses regression methods like CART, Tree Bagger, Linear Regression, Gaussian Process Regression (GPR) and Support Vector Regression (SVR). The prediction accuracies of these methods are compared and K-mean clustering technique is implemented to improve the prediction accuracy. The suitability of these models has also been discussed in details in this thesis.||URI:||http://hdl.handle.net/10356/73143||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.