Combined Gaussian Mixture Model and Pathfinder Algorithm for Data Clustering.

Area of Science:

• Machine learning and data analysis
• Artificial intelligence
• Computational statistics

Background:

• Gaussian Mixture Models (GMMs) are widely used for data clustering but require manual cluster number specification and can suffer from poor initialization.
• Limitations of GMMs include sensitivity to initial parameters and the inability to automatically determine the optimal number of clusters.

Purpose of the Study:

• To introduce a novel clustering algorithm, PFA-GMM, designed to overcome the limitations of traditional GMMs.
• To enable automatic determination of the optimal number of clusters.
• To enhance the initialization process and avoid local convergence issues in clustering.

Main Methods:

• The proposed PFA-GMM algorithm integrates the Pathfinder algorithm (PFA) with Gaussian Mixture Models (GMMs).
• PFA is utilized to guide the initialization process and determine the optimal number of clusters.
• The algorithm treats clustering as a global optimization problem to mitigate issues with local convergence.

Main Results:

• Comparative studies using synthetic and real-world datasets demonstrate the effectiveness of PFA-GMM.
• PFA-GMM significantly outperformed established clustering algorithms in terms of accuracy and efficiency.
• The algorithm successfully automated cluster number determination and improved initialization robustness.

Conclusions:

• PFA-GMM offers a robust and automated solution for data clustering, addressing key limitations of standard GMMs.
• The integration of PFA enhances GMM performance by optimizing cluster number selection and initialization.
• This novel approach represents a significant advancement in machine learning for data analysis.