CBRW - Detecting Outliers in Non-IID Categorical Data
The Coupled Biased Random Walk model (CBRW for short) is an outlier detection method that models a complex intra- and inter-feature value couplings or interactions to estimate the outlier scores of feature values in categorical data sets. The value outlier scores are then integrated to compute the outlier scores of data objects or features to facilitate outlier detection, or outlying feature selection (feature selection for subsequent outlier detection methods). CBRW is able to handle non-IID categorical data, i.e., data with heterogeneous feature values and complex value couplings due to its capability in capturing the complex value interactions.
The source code of CBRW and its executable jar file is available at here.