Multi-Resolution Spatial Outlier Detection (MRSOD)

Outlier detection focuses on identifying abnormal patterns from large data sets. In this thesis, we propose a novel approach for outlier detection called Multi-Resolution Spatial Outlier Detection (MRSOD) to identify spatial outliers by considering multiple levels of attributes in complex spatio-temporal datasets. Current spatial data mining techniques have primarily focused on single-resolution data where the data is at one level such that every attribute is simple and all records have the same number of attributes. However in reality spatio-temporal data is often complex such that an attribute may have additional levels of details associated with it and different spatial objects may have different attributes associated with them. We present detailed experimental results in Maryland Water Monitoring Sensors data from U.S. Geological Survey and simulated highway sensor data. Our results indicate that the MRSOD approach finds outliers with high accuracy in both the datasets