Multi-Dementional Feature Engineering for Reliable Detection Method of Vehicles with Location Information Errors in Spatio-Temporal Data Retention System

The widespread adoption of IoT devices has led to increased sensor data. Among the data, we defined spatiotemporal data (STD) as data that depends on both time and space. We proposed a spatiotemporal data retention system (STD-RS) that uses vehicles to retain and utilize STD within a specific spatial area. However, in STD-RS, errors in vehicle location information can cause data to be transmitted outside the target area. Previously, we proposed a method to detect abnormal vehicles by learning the distance and received signal strength indication (RSSI) of the vehicle and then confirmed that the method can detect abnormal vehicles with an accuracy of approximately 80%. However, the previous method did not consider vehicles that cannot generate features for learning in this paper. Moreover, as the evaluation was conducted in a single (ideal) environment only, the validation of effectiveness in practical environments remains. In this study, we use the sufficiency rate of data necessary for feature generation as an evaluation metric and evaluate how the spatial and quantitative changes in the combination of features for learning impact the sufficiency rate and detection accuracy under practical environments. The simulation evaluation results revealed the factors that affected the sufficiency rate and demonstrated that using the long-term spatial variation of RSSI as a feature improves the detection accuracy in any environment.