Digit Forest for Outlier Detection

Outlier detection plays a vital role in modern computer science due to its important applications such as noise removal, fraud detection, and modern industrial monitoring. Although many outlier detection methods have been created over the years, only a small number of them effectively exhibit a good balance between high performance accuracy and low computational complexity. Owing to this drawback, the usage in large-scale data applications is therefore limited. To tackle this challenge, a novel method called Digit Forest (DF) is proposed in this paper. Furthermore, DF operates on the same principle as the isolation forest (IF) outlier detector in which outliers with lower density are identified. Moreover, DF exhibits lower complexity compared to IF as the dataset size increases. Experimental results indicate that DF surpasses all 16 baseline methods tested on 8 publicly available real-world datasets.