Using Machine Learning to Improve Interactive Visualizations for Large Collected Traffic Detector Data

In traffic engineering, cities rely on large detector datasets to manage traffic. Visualizing these big, multi-dimensional datasets poses challenges such as overplotting and dimension reduction, often rendering traditional visualization techniques inadequate. To address this, we added two machine learning (ML) algorithms (Local Outlier Factor algorithm and K-Prototypes clustering) to an interactive time series visualization to improve exploration by both domain experts and non-experts. We used an original detector dataset of a mid-sized German city. Our findings reveal that the ML algorithms greatly enhanced data exploration in these interactive visualizations, particularly for users with limited domain knowledge. This research directly contributes to the design of traffic data analysis tools, offering a foundation for traffic detection hardware and software improvements but also advancing complex dataset visualization in general. It will ultimately lead to more informed decisions, improved traffic management, and has the potential to reduce air pollutants, thus counteracting climate change.