Anomaly Detection and Imputation for Tartu Traffic Sensors
Name
Joonas Praks
Abstract
The city of Tartu has 16 highway traffic sensors with many gaps of missing data. We analyzed the state of the sensors’ data and evaluated different anomaly detection and imputation solutions to better its quality. The best anomaly detection approach was deemed to be daily clustering with local outlier factor (LOF) used as the clustering algorithm. For imputation we utilized linear interpolation with a combination of seasonal decomposition and seasonal splitting. The chosen solutions were integrated into a service that processes CSV files of traffic data and uploads the results to Cumulocity, an IoT data aggregation platform. We processed and uploaded the historical data of 2019-04-29 to 2023-06-01 of every highway sensor. Finally, we also tested our solution on light traffic data.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Pelle Jakovits
Defence year
2024