Inference of vehicular traffic in smart cities using machine learning with the internet of things

A rapidly increasing world population is not only directly related with increased urban mobility but also traffic delays, losses in fuel, air and noise pollution as well as physiological and psychological problems in the population. Traffic control has been shown to be a means of maintaining road serviceability with a key input being traffic inference. The harvesting of sufficiently large data sets for traffic inference has thus far been a challenge examined by the scientific community. A modelling, interactive design approach to the harvesting and preprocessing of vehicular data can be used to bridge the gap for traffic control to be autonomously implementable in an urban context. With this consideration, this paper explores the implementation of an internet of things multiagent system, in a network and equipped with sensors, distributed along the roadway to collect and share this vehicular data among its nodes and then process the data using a machine learning algorithm for learning and subsequent inference of vehicular types. An interactive design approach is used in the creation of a reduced model of knowledge from each node’s bulk harvested data, which is then easily stored or deployed and which also facilitates post-processing of data. Considering the smart city as a product, the interactive approach to product engineering permits the implementation of these mathematical models as part of an interactive traffic control system, to ensure the city’s adaptability to the ever-changing needs of its inhabitants. In this paper, we present the analysis of the trivial case of the detection of presence and absence of a class A type vehicle. In the introductory section, we examine the problem of transportation in urban expansion. In section two, we examine the solution provided by the Smart Cities with respect to transport services. Further problems which appear within these transport services and their respective solutions in machine learning and the internet of things are discussed in this section on these disruptive technologies. Section three examines the bibliography for successes achieved with similar research and also demonstrates the basis for classification by support vector machines. Sections four to six explain how the experiment is set up to test the hypothesis that over 99% accuracy in vehicular detection can be achieved with the proposed system when cross-referenced with the actual acoustic, magnetic and video presence and absence of a test vehicle.