Anomaly detection refers to the process of finding outlying records from a given dataset.This process is a subject of increasing interest among analysts. Anomaly detection is a subject of interest in various knowledge domains. As the size of data is doubling every three years there is a need to detect anomalies in large datasets as fast as possible. Another need is the availability of unsupervised methods for the same. This thesis aims at implement and comparing few of the state of art unsupervised outlier detection methods and propose a way to better them. This thesis goes in depth about the implementation and analysis of outlier detection algorithms such as Local Outlier Factor(LOF),Connectivity-Based Outlier Factor(COF),Local Distance-Based Outlier Factor and Influenced Outlierness. The concepts of these methods are then combined to propose a new method which better the previous mentioned ones in terms of speed and accuracy.
展开▼
