Multilayer Perceptron and Particle Swarm Optimization Applied to Traffic Flow Prediction on Smart Cities

Abstract

Smart cities can increase the live quality of its citizens and Intelligent Transportation Systems is a key topic in this area. When the population density living in the same region increases more and more, the cities suffer from problems as constant traffic jams. Thinking this way, in this paper are present the uses of computational intelligence techniques and analyses to aid in traffic dimensioning solutions. To do this, prediction models and heuristics are the best way to create a more autonomous and intelligent environment. In this work, an application is introduced applying machine learning and an optimization technique to empower a smart ecosystem. To validate it, an evaluation using Multi-Layer Perceptron together with Particle Swarm Optimization was performed, comparing it with the state-of-the-art. All evaluations were done using real data traffic with a free traffic flow scenario. Applying the Particle Swarm Optimization to optimize the activation functions’ parameters, we obtained 3.1% average MAPE for Logistic activation function and a MAPE of 3.4% for ReLU activation function.