Outlier Detection and Explanation Method Based on FOLOF Algorithm.

Area of Science:

• Data Science
• Machine Learning
• Anomaly Detection

Background:

• Conventional outlier detection methods struggle with heterogeneous data and high computational costs.
• Existing algorithms often lack essential data preprocessing steps, limiting their effectiveness.

Purpose of the Study:

• To introduce FOLOF (FCM Objective Function-based LOF), an improved local outlier detection method.
• To address limitations of existing outlier mining techniques, including data preprocessing and computational efficiency.

Main Methods:

• Utilizing the elbow rule to determine optimal data clusters.
• Employing the FCM objective function for initial outlier candidate pruning.
• Applying a weighted local outlier factor algorithm for anomaly scoring.
• Using the Golden Section method for outlier classification.

Main Results:

• FOLOF demonstrates effectiveness in identifying outliers across artificial, UCI, and NBA player datasets.
• The method successfully prunes datasets to identify candidate outliers efficiently.
• Analysis of outlier factors per dimension reveals underlying causes of anomalies.

Conclusions:

• FOLOF offers a robust and computationally efficient approach to outlier mining.
• The method enhances outlier detection performance by incorporating data preprocessing.
• FOLOF provides insights into the characteristics of anomalous data points.