International Journal of Computer Applications (0975 – 8887) Volume 183 – No. 27, September 2021 10 Design and Evaluation of a New Machine Learning Toolbox for Optimal Traffic Light Control with SUMO and Tensorflow Reda Mali LTI Laboratory, National School of Applied Sciences, Chouaib Doukkali University El Jadida, Morocco Mohammed Bousmah LTI Laboratory, National School of Applied Sciences, Chouaib Doukkali University El Jadida, Morocco ABSTRACT Today, all the major metropolises of the world suffer from serious problems of congestion and saturation of road infrastructures. Within this context, one of the main challenges is the creation of appropriate Machine Learning algorithms for the optimization of the traffic lights systems. The objective is to minimize the total journey time of the vehicles that are present in a certain part of a city. In this article, we propose a new toolbox and a framework that brings Tensorflow features to Simulation of Urban Mobility (SUMO). Our work aims to facilitate the use of the SUMO simulator with Tensorflow, for road traffic management. With this tool, researchers will be able to easily test their different models quickly. Instead of spending several days studying the SUMO API, and setting up data mapping procedures, researchers will be able to get results in minutes with our tool. A Web generator let researchers set simulation scenarios, and they can implement their model with the toolbox, based on neural networks and Deep Q Learning. The toolbox exports many metrics, and can compare multiple policies, and different hyper parameters to optimize models. The experimental results obtained show that such an approach makes it possible to obtain significant gains. Keywords Traffic Light Control; Machine Learning; Simulation Tool; Deep Q Learning. 1. INTRODUCTION Nowadays, the road infrastructure is under pressure. This is mainly due to the increase in the number of private vehicles, and the difficulty of creating new infrastructure or upgrading existing one. Optimizing the use of road infrastructure is therefore becoming a major issue [1]. However, it is difficult to implement a manual solution, as a large number of scenarios must be considered. It is also necessary to consider the case of special vehicles, accidents, weather conditions, etc. Current traffic management systems are based on predefined rules, and usually require human intervention to work optimally [2]. Our goal is to democratize the implementation of artificial intelligence-based solutions to solve this problem. With the advent of smart cities, and the democratization of object recognition and tracking tools, it is possible to implement intelligent solutions based on artificial intelligence, which can be deployed on a large scale. Researchers have at their disposal several simulation tools, such as SUMO (Simulation of Urban Mobility), which allow to create very detailed simulations [3]. They also have very powerful tools, such as Tensorflow, which allows to create models based on deep learning and reinforcement learning. However, it is difficult to master these tools, and the learning curve is usually long. Our work consists in proposing a solution that integrates the power of both SUMO and Tensorflow. With this, researchers will be able to focus mainly on models. We propose a set of tools and a framework, which allows to test a model in a few minutes, and to take advantage of the power of SUMO and Tensorflow immediately. Our paper is organized as follows: Section 2 presents the background of intelligent traffic management. Section 3 details the design of our tool, and its different components. And finally, section 4 presents a use case provided with the tool, and the results that the tool can provide. 2. BACKGROUND 2.1 IA and Smart cities Today, cities should face several challenges, such as population growth, energy consumption, environment preservation, life quality life improvement for citizens, etc. These challenges require the implementation of new techniques and solutions. In the begin of this century, the role of Information and Communication Technologies (ICT) to improve transportation in smart cities has been shown. In this section , we will discuss the role of Artificial Intelligence (IA) in smart cities. Indeed, with the emergence of Big Data, Internet of Things (IoTs) and 5G, AI can propose innovative and optimal policies to solve smart city problems. Techniques such as machine learning, deep learning, and reinforcement learning consider the specificities of each city in order to implement effective management solutions and policies. In this section, we will present several aspects of AI and their potential in smart cities [4]. 2.2 Machine Learning Machine Learning provides algorithms that automatically learn from past experiences. This field of research has seen many advances in recent years [5]. These advances are achieved due to new algorithms and the availability of online data. Machine learning regained popularity, thanks to the results obtained in the field of image recognition. Subsequently, scientists have felt the potential of machine learning and it’s used in other problems such as natural language processing, speech recognition, traffic prediction, fraud detection, etc. Its strength comes from the fact that it is sometimes more efficient to solve problems using learning, rather than to implement solutions manually.