Outlier detection using iterative adaptive mini-minimum spanning tree generation with applications on medical data.

Area of Science:

• Data Science
• Machine Learning
• Medical Informatics

Background:

• Outlier detection is vital for data pre-processing, particularly in medical applications.
• Existing methods often fail with varying outlier distributions or require predefined outlier proportions.
• This limitation hinders the reliable application of outlier detection techniques.

Purpose of the Study:

• To propose a novel outlier detection method that overcomes limitations of existing approaches.
• To develop an adaptive technique that does not require prior knowledge of outlier percentages.
• To enhance the robustness of outlier detection across datasets with varying densities and shapes.

Main Methods:

• An adaptive mini-minimum spanning tree-based outlier detection (MMOD) method was developed.
• A novel distance measure, scaling Euclidean distance, was utilized.
• The method constructs a minimum spanning tree to identify outliers adaptively.

Main Results:

• MMOD demonstrated effectiveness in identifying outliers across datasets with different densities and shapes.
• The method successfully detected outliers without prior knowledge of their proportion.
• Performance was validated on both synthetic and real-world medical datasets.

Conclusions:

• The proposed MMOD method offers a robust and adaptive solution for outlier detection.
• It addresses key limitations of existing techniques, particularly the need for prior outlier percentage knowledge.
• MMOD shows significant potential for application in medical data analysis and other fields.