Density-Based Outlier Detection: Supervised Approach Based on Virtual Points

- Outlier detection plays an important role in a wide variety of domains and has many applications. The examples are fraud detection, anomaly detection in industrial systems, healthcare, cybersecurity, environmental monitoring and marketing. The aim of outlier detection is to identify and analyze data points or observations that differ significantly from the rest of the data that may indicate potential fraud, disease outbreaks, or equipment failures. The unsupervised approach to outlier detection is often preferred because it does not require labeled data, meaning that the algorithm can identify outliers in a dataset without having to know beforehand which observations are anomalous. This is particularly useful when there is limited prior knowledge about the data, or when the anomalous observations are rare or unknown. We propose a novel approach to outlier detection which doesn't require labeled data but is based on the supervised learning technique similar to decision tree construction. Our method overcomes many of the limitations of traditional outlier detection techniques by partitioning the space into cluster and empty regions using a decision tree. We accomplish this by introducing virtual data points and modifying the decision tree algorithm accordingly. The results of our experiments on synthetic and real-world datasets show that the method is both highly efficient and scalable.