Distance-based k-nearest neighbors outlier detection method in large-scale traffic data

This paper presents a k-nearest neighbors (kNN) method to detect outliers in large-scale traffic data collected daily in every modern city. Outliers include hardware and data errors as well as abnormal traffic behaviors. The proposed kNN method detects outliers by exploiting the relationship among neighborhoods in data points. The farther a data point is beyond its neighbors, the more possible the data is an outlier. Traffic data here was recorded in a video format, and converted to spatial-temporal (ST) traffic signals by statistics. The ST signals are then transformed to a two-dimensional (2D) (x, y) -coordinate plane by Principal Component Analysis (PCA) for dimension reduction. The distance-based kNN method is evaluated by unsupervised and semi-supervised approaches. The semi-supervised approach reaches 96.19% accuracy.