OUTLIER DETECTION WITH DATA STREAM MINING APPROACH IN HIGHDIMENSIONAL TIME SERIES DATA By WANG DAN TONG Thesis Supervisor: Dr. Simon Fong Department of Computer and Information Science Master of Science in E-commerce Technology Outlier detection has applications in fraud identification in the industries. It can also serve as a preprocessing technique that cleans noise from the training data for the subsequent model induction. As a preprocessing technique, it is necessary to distinguish whether the outliers are superfluous noises which warrant removal, or they are interesting anomalies which should be learnt as well. In this thesis, the author presents a novel algorithm that uses probabilistic and density-based model combined with incrementally optimized very fast decision tree (ioVFDT) for embracing outliers and for generalizing recognition powers. The outlier removal mechanism is empowered by logics that keep counts of outlier occurrence. Additional classification rules are formulated for outliers that consistently map data to correct target classes. Otherwise they would be removed. The proposed algorithm is tested with nine different scenarios time series datasets for verifying its efficacy.