Detection of Anomalies in Traffic Flows with Large Amounts of Missing Data

Anomaly detection plays an important role in traffic operations and control. Missingness spatial-temporal datasets prohibits anomaly algorithms from learning characteristic rules patterns due to the lack of large amounts data. This paper proposes scheme for 2021 Algorithms Threat Detection (ATD) challenge based on Gaussian process models that generate features used a logistic regression model which leads high prediction accuracy sparse flow data with proportion missingness. The dataset is provided by National Science Foundation (NSF) conjunction Geospatial-Intelligence Agency (NGA), it consists thousands labeled records 400 sensors 2011 2020. Each sensor purposely downsampled NSF NGA order simulate missing completely at random, rates are 99%, 98%, 95%, 90%. Hence, challenging detect anomalies proposed makes use different times day days week recover complete computationally efficient allowing parallel computation sensors. method one two top performing ATD challenge.