Empirical Analysis of Traffic patterns: Leveraging Machine Learning Techniques in-Depth Insights

Efficient traffic modeling and control depend on accurate travel condition data. Previously, limited data forced analysts to use single days or small averages, missing minor variations. Now, increased data improves accuracy but complicates traffic monitoring, as daily model calibration is impractical. This paper presents a systematic methodology that uses machine learning-based cluster analysis to categorize days into distinct groups representing specific traffic patterns providing more precise responses algin with temporal variation travel conditions. Instead of pre-selecting groups, clustering discovers natural groups that exist within the data. Initially, traffic time series data over eleven months from a motorway section in Italy were cleaned and processed against missing data and noises. Subsequently, three clustering approaches, K-Means, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and Affinity Propagation (AP) were used to identify groups with similar traffic patterns and the performance of different approaches are evaluated. Results confirm the compatibility of the K-Means and DBSCAN algorithm effectiveness. Conversely, AP suggests a significantly higher number of clusters, with negligible differences. As expected, notable distinction is noted by clusters between the school period, non-school period, weekdays, and weekends. The usage of clustering algorithms facilitated the identification of six distinct day types within the dataset. Furthermore, this approach can be applied daily to detect outliers and anomalies in both demand and supply. This is a noticeable feature to support mobility agencies in planning and applying special policies to face anomalies conditions.