In the bustling hubs of our cities, every second counts when it comes to detecting and responding to traffic incidents. Imagine a world where the time between a traffic incident occurring and it being resolved is minimized to the bare essentials. Urban congestion and traffic incidents are more than mere inconveniences; they are significant societal issues that escalate greenhouse gas emissions and impact the quality of life of citizens. These incidents, which include anything from vehicle breakdowns to accidents and collisions, not only disrupt the flow of traffic but also lead to delays, property damage, and worse, injuries and fatalities.
Addressing these challenges requires innovative solutions, and this is where our work within the FRONTIER project introduces a novel approach for traffic incident detection, by harnessing the ever growing use of Automated Machine Learning (AutoML).
What is AutoML, one might ask? AutoML is like a smart assistant for data tasks. It takes the complex steps of building a machine learning model – which usually involve a lot of technical know-how – and makes them automatic. [1] Think of it as a helper that can pick the best tools for understanding data, use them, and keep getting better over time, all by itself. Setting it up might need some extra computer power at the start, but it saves a lot of time and effort later. That means more people can use machine learning, even if they're not experts. intervention [2]. With AutoML, businesses and individuals can make smart use of their data without needing to learn all the details. It's all about making the benefits of machine learning easy and accessible for everyone.
ICCS is the lead partner within FRONTIER project responsible for developing solutions for automatic detection of emerging situations and incidents. Our recent research takes a dive into this innovative realm which has not been extensively explored within the transportation field. Traditional methods of incident detection in Intelligent Transport Systems (ITS) rely on set patterns and rules, but they often fall short in the face of unpredictable, non-recurrent traffic incidents. Here's where Machine Learning (ML) has been changing the game, providing flexible and adaptive solutions, and now AutoML lies at the heart of modern data-driven solutions, as a transformative technology designed to streamline and simplify the ML workflow.
Our methodology utilizes the Tree-based Pipeline Optimization Tool (TPOT) to streamline the transition from raw data to practical predictions. Tree-based Pipeline Optimization Tool (TPOT) is a machine-learning library that requires few lines of code and makes the machine-learning processes more streamlined by automating tasks such as data preparation, model selection, hyperparameter tuning, and deployment. [3] Our methodology employs a range of ML techniques and optimization algorithms, from both a classification and regression standpoint, ultimately enhancing the accuracy and efficiency of the incident detection process.
Our real-world tests in the cities of Athens, Greece, and Antwerp, Belgium, have shown promising results. When we compare our AutoML-based approach against general-purpose ML algorithms, the former displayed impressive efficiency based on the established metrics. Our methodology isn't just about improved performance—it's about transforming urban mobility management. By simplifying the process, we aim to inspire more cities to embrace intelligent traffic systems, leading to safer and more fluid urban travel. Our work doesn't shy away from the limitations of our approach; yet, we also point to the exciting possibilities that AutoML presents, not just in traffic management but across various domains where predictive analytics play a vital role.
As we navigate the evolving landscape of urban transport, we're grateful to contribute to the collective efforts with technologies like AutoML. Keep an eye on our progress as we continue to explore and expand the boundaries of intelligent transportation solutions!
Works Cited
[1] F. Hutter, L. Kottho and J. V. (Eds.), Automated Machine Learning: Methods, Systems, Challenges, Springer, 2018.
[2] H. Song, I. Triguero and E. . Ozcan, "A review on the self and dual interactions between machine learning and optimisation," Progress in Artificial Intelligence , pp. 1-23, 2019.
[3] R. Olson, N. Bartley, R. Urbanowicz and J. Moore, "Evaluation of a tree-based pipeline optimization tool for automating data science," in Proceedings of the Genetic and Evolutionary Computation Conference 2016, New York, 2016.
Authors: G.Gkioka, G.Mentzas (ICCS-NTUA)