An Improved Outlier Detection Algorithm Based on Distance Increment and Decision Graph Score

Abstract

In recent years, outlier detection technology has played an important role in all aspects of real life. Faced with increasingly complex production datasets, how to find rare outliers quickly and efficiently is a huge challenge. In this article, we introduce an improved LOF algorithm which is based on increment of distance and decision graph score. Firstly, we use distance increment to calculate the local density of samples. Then we use decision graph score to quantify the anomaly degree of samples, which is the ratio of distance to local density. The higher the score, the more likely it is to be an outlier. Experiments show that this method has a good performance on both synthetic datasets and real datasets, and can effectively deal with complex datasets.