Data Analytics for Smart Parking Applications.

Area of Science:

• Computer Science
• Data Science
• Smart City Technologies

Background:

• Smart parking systems collect real-time occupancy data from sensors.
• Ensuring data quality is crucial for effective parking management and citizen services.
• Existing methods for sensor data analysis have limitations in accuracy and outlier detection.

Purpose of the Study:

• To develop and validate an automated classification algorithm for smart parking sensor data.
• To detect anomalous sensor behavior (outliers) and group sensors with similar patterns (clustering).
• To enhance the utility of parking data for end-users like parking managers and citizens.

Main Methods:

• Analysis of real-world smart parking data statistics to create simulation models.
• Development of a sophisticated algorithm using unsupervised learning techniques, specifically self-organizing maps (SOMs).
• Supervised tuning of the SOM algorithm using a generated trace dataset and comparison with Expectation Maximization, k-means, and DBSCAN.

Main Results:

• The proposed self-organizing maps algorithm demonstrated superior classification accuracy compared to other clustering methods.
• The algorithm successfully identified all outlier sensors in a six-month real-world dataset.
• The approach effectively groups parking sensors exhibiting similar occupancy patterns.

Conclusions:

• The developed unsupervised learning algorithm significantly improves the reliability and accuracy of smart parking data.
• This method enhances the practical application of smart parking data for operational efficiency and user experience.
• The algorithm provides a robust solution for outlier detection and sensor data clustering in intelligent transportation systems.