A hybrid distance-based outlier detection approach

Most real-world datasets have outliers. Outliers can imply abnormal states that often indicate significant performance degradation or danger in certain circumstances. Therefore, the outlier detection plays an important role in the field of data mining. This paper proposes a hybrid distance-based outlier detection approach. It uses the average distance as neighborhood distance, and records the number of data objects within the neighborhood. Therefore, the average number of neighbors can be calculated. Using this average value as a threshold, the data set can be divided into two parts: non-outlier data set and candidate data set. Calculating the distances between a candidate object and its k-nearest neighbors can filter out outliers from the candidate data set. Experimental results show that the approach can effectively detect outliers.