DEVELOPMENT OF A DEEP LEARNING MODEL FOR PROFILING AND PREDICTING TRAFFIC OFFENDERS

Abstract

Traffic violation has been a big challenge to mankind as many of the accidents recorded over years is as a result of traffic violation. The traffic violation monitoring systems deployed in developing countries lack adequate authentication of road users, no efficient and effective profiling system of offenders and offences, as such, traffic agents cannot generate instant traffic history of an offender. The monitoring of traffic offenders in developing countries has a lot of challenges including; lack of proper authentication of vehicles and users, lack of substantive traffic system that suits the management of traffic offenders’ profile in both rural and urban areas, lack of predictable modules to forecast the tendency of an offender to cause accident in the future, poor means of communication between traffic agencies and vehicle users, poor traffic offence awareness for vehicle users and lack of a dependable traffic offenders profile database. The area of traffic offender identification using fingerprint and NIN and predicting the possibility of the traffic offender committing more traffic offence in the nearest future has not be researched on and this forms the research gap this paper is set to cover. Therefore, this thesis is providing a solution by development of a deep learning model for profiling and predicting traffic offender’sfocuses on developing a traffic offenders profiling and prediction system using deep learning algorithm to predict the likelihood of an offence to be committed by a road user.The proposed system developed a model that will profile traffic offenders in both urban and rural settings, create a traffic offender’s database that will interact with existing national databases to authenticate traffic offenders, provides a module that will predict the likelihood of a road user to commit severe traffic blunder in the future and provide intelligent information necessary for timely action by law enforcement agencies. The system also createdan SMS based traffic awareness module that handles traffic offences communication between traffic agents and offenders. These designs are implemented using a websystem developed with PHP, MySQL and JavaScript. The System Design followed the OODM methodology for componentization of the system modules giving room for coupling, decoupling, modification, encapsulation and reuse, as well as easy maintainability. Unified Modeling Language was extensively used to simplify the explanation of the system modules. The software performance was tested using accuracy of traffic offender prediction and Confusion Matrix was adopted for the thesis. For the purpose of this thesis, dataset was collected from data-world (https:// data. world/ health data ny/qutr-irdf) and the data is an excel sheet dataset for Traffic offenders. The result obtained from the new system developed shows 95% accuracy of the deep learning technique for predict the likelihood of a road user to commit traffic blunder in the future.