Path Normalcy Analysis Using Nearest Neighbor Outlier Detection.
David Luper, Muthukumaran Chandrasekaran, Khaled Rasheed, Hamid Reza Arabnia · International Conference on Artificial Intelligence · 2008
We present a machine learning technique that recognizes patterns of normal movement, using GPS data and time stamps, to gain the ability to detect regions of time containing abnormal movement. We argue people move throughout regions of time in established patterns, and a person’s normal movement can be learned by machines. We use intelligent features extracted from raw GPS data with time stamps, to describe a person’s movement over discrete regions of time. Then we use a nearest neighbor approach to determine outliers in a distribution of time regions. We consider outliers as time regions where patterns of established normal movement have been violated. Ultimately, we produce a distance range value for a distribution in conjunction with normalized scores depicting the degree to which each time region contained movement consistent with the other time regions being analyzed. We also produce a classification of each day as normal or abnormal.