| This thesis studies the properties of distance-based outliers and a better detection method for large multi-dimensional datasets. Outlier detection is an important task to find out the objects that deviate in a high ratio from the rest of the objects. The proposed algorithm breaks the data set into divisions and sets the area of access for each division, thus reducing the unnecessary access for a major set of elements. This algorithm reduces the run time of the existing algorithm by using separators. Datasets of varying sizes have been tested to analyze the empirical values of these procedures. Effective data structures have been implemented to gain efficiency in memory-performance. |
