Detection of outliers in high-dimensional data using nu-support vector regression.

Area of Science:

• Machine Learning
• Data Mining
• Computational Statistics

Background:

• High-dimensional data (HDD) presents challenges in outlier detection and classification.
• Accurate outlier identification is crucial for reliable model fitting and interpretation in applied sciences.
• Existing methods like standard Support Vector Regression (SVR) and mu-epsilon-SVR are computationally expensive for outlier detection.

Purpose of the Study:

• To propose and evaluate nu-Support Vector Regression (nu-SVR) as an efficient method for outlier detection in high-dimensional data.
• To address the computational limitations of existing SVR-based outlier detection techniques.
• To demonstrate the effectiveness of nu-SVR in identifying outliers across various scenarios.

Main Methods:

• Implementation of nu-Support Vector Regression (nu-SVR) for outlier detection.
• Evaluation of the proposed nu-SVR method using three real-world datasets.
• Validation through Monte Carlo simulations to assess performance under diverse conditions.

Main Results:

• The proposed nu-SVR method demonstrates high success rates in identifying outliers in high-dimensional data.
• nu-SVR significantly reduces computational running time compared to standard and mu-epsilon-SVR methods.
• The method's efficacy is robust across different datasets and simulation settings.

Conclusions:

• nu-Support Vector Regression (nu-SVR) is a highly effective and computationally efficient technique for outlier detection in high-dimensional data.
• The proposed nu-SVR approach provides a valuable alternative for applications requiring robust outlier identification.
• Further research can explore nu-SVR in specific domains like medical diagnostics for cancer cell classification.