Robust Random Walk Based on Natural Neighbors for Outlier Detection

Area of Science:

• Data Science
• Machine Learning
• Anomaly Detection

Background:

• Outlier detection is crucial for identifying anomalous data points in complex datasets.
• Existing methods struggle with noise and boundary samples, degrading relationship quality.
• Traditional random walk methods are susceptible to weak connections, leading to inaccurate results.

Purpose of the Study:

• To propose a robust random walk based on natural neighbors for outlier detection (RWNOD).
• To enhance outlier detection by mitigating noise and preserving structural information.
• To improve the reliability of random walk approaches in identifying anomalous samples.

Main Methods:

• An adaptive smoothing mechanism using natural neighbors to adjust sample positions and reduce noise.
• A robust random walk strategy incorporating shadowed sets into the transition matrix to preserve reliable connections.
• Development of a corresponding outlier detection algorithm.

Main Results:

• The proposed RWNOD method demonstrates superior performance in outlier detection.
• Experimental results confirm the strong robustness of the RWNOD algorithm.
• The method effectively handles noise and boundary samples, improving relationship quality.

Conclusions:

• The RWNOD method offers a significant advancement in robust outlier detection.
• Leveraging natural neighbors and shadowed sets enhances the accuracy and reliability of random walk-based anomaly detection.
• The proposed approach provides a more effective solution for identifying anomalous samples in complex data.