Prediction of Average Speed Based on Relationships Between Neighbouring Roads Using K-NN and Neural Network

Bagus Priambodo,Rabiah Abdul Kadir,Azlina Ahmad

doi:10.3991/ijoe.v16i01.11671

Abstract

For decades, various algorithms to predict traffic flow have been developed to address traffic congestion. Traffic congestion or traffic jam occurs as a ripple effect from a road congestion in the neighbouring area. Previous research shows that there is a spatial correlation between traffic flow in neighbouring roads. Similar traffic pattern is observed between roads in a neighbouring area with respect to day and time. Currently, time series models and neural network models are widely applied to predict traffic flow and traffic congestion based on historical data. However, studies on relationships between road segments in a neighbouring area are still limited. It is important to investigate these relationships because they can assist drivers in avoiding roads which are impacted by road congestion. Also, the result can be used to improve the accuracy of prediction of traffic flow. Hence, this study investigates relationships of roads in a neighbouring area based on similarity of traffic condition. Traffic condition is influenced by number of vehicles and average speed of vehicles. In our study, clustering method is used to divide the speed of traffic into four (4) categories: very congested, congested, clear and very clear. We used k-means clustering method to cluster condition of traffic flow on road segments. Then, we applied the k-Nearest Neighbour (k-NN) method to classify the traffic condition in neighbouring roads. From the classification of traffic condition in neighbouring roads, we then determine the relationship between road segments. We presented the road with highest relationship on the map and used it as input factor to predict traffic speed of the road using neural network. Results show that combination of k-means and k-NN method produced better results than using both, correlation method and using the k-means method only.

Highlights

Due to increase in population and number of private cars in this modern era, traffic congestion has become significantly worse, leading to economic losses and causing human stress[1] and environmental damages such as pollution [2]
Instead of using the k-Nearest Neighbour (k-NN) method for regression, we proposed k-NN to classify traffic conditions in neighbouring roads based on results from clustering of traffic conditions of a particular road, since the k-NN classifier is considered as the most popular classifier for pattern recognition with its simplicity [23]
Clustering road destination based on traffic condition: Traffic flow condition is influenced by traffic speed of vehicles passing through and the number of vehicles [21]

Summary

Introduction

Due to increase in population and number of private cars in this modern era, traffic congestion has become significantly worse, leading to economic losses and causing human stress[1] and environmental damages such as pollution [2]. Availability of traffic information can lead to chain changes in traffic flow condition in upstream and downstream of road segments in the same area. Traffic congestion is a circumstance in which road users exceed the capacity of the road. Effective methods or models need to be developed to identify congested links, to analyses the relationship between the occurrence of congestion and increasing traffic flow, and to find congestion distribution in the network. Some studies use vehicle speeds, weather, incident, and special days to predict traffic flow using linear regression[4] and neural network models[5][6]. Other factors that influence traffic congestion that they are dynamic and interrelated. Traffic congestion or traffic jams can propagate from one road to other roads in neighbouring roads [7]

Methods

Discussion

Conclusion