Estimation of Radial Basis Function Network Centers via Information Forces.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Determining optimal centers for Radial Basis Function Networks (RBFN) remains a significant challenge in machine learning.
• Effective clustering and outlier detection are crucial for accurate data classification and analysis.

Purpose of the Study:

• To propose and evaluate a novel gradient algorithm for determining RBFN cluster centers using information forces.
• To develop a robust method for outlier classification based on Information Potential.
• To assess the performance of the proposed algorithms on diverse and challenging datasets.

Main Methods:

• A gradient algorithm is employed to identify cluster centers based on information forces acting on data points.
• Information Potential is utilized to establish a threshold for classifying outliers.
• The proposed RBFN approach is benchmarked against a standard k-means clustering algorithm.

Main Results:

• The proposed gradient algorithm successfully determines cluster centers for RBFN.
• The Information Potential threshold effectively identifies and classifies outliers.
• The combined approach demonstrates superior performance compared to RBFN with k-means, especially on datasets with varying cluster sizes, overlap, and noise.

Conclusions:

• The information force-driven gradient algorithm offers an effective solution for RBFN center determination.
• The developed outlier classification method enhances the robustness of RBFN.
• This approach provides a competitive alternative for data classification tasks, outperforming traditional methods in complex scenarios.