A Novel BN Learning Algorithm Based on Block Learning Strategy.

Area of Science:

• Computational statistics
• Machine learning
• Data mining

Background:

• Learning Bayesian Network (BN) structures from high-dimensional and sparse data presents significant computational challenges.
• Existing methods often struggle with scalability and efficiency when dealing with complex datasets.

Purpose of the Study:

• To develop a more efficient algorithm for accurate Bayesian Network structure learning in high-dimensional and sparse data.
• To address the high computational complexity associated with traditional BN learning methods.

Main Methods:

• Proposes a block learning algorithm with a mutual information-based K-means algorithm (BLMKM).
• Employs an improved K-means algorithm for node blocking and a maximum minimum parents and children (MMPC) algorithm for skeleton identification.
• Utilizes pruned dynamic programming and a scoring function to identify the optimal Bayesian Network structure.

Main Results:

• The BLMKM algorithm achieves similar accuracy to non-blocking classical algorithms in a reasonable time for high-dimensional, sparse data.
• Demonstrates a time advantage over existing block learning algorithms while maintaining accuracy.
• Successfully applied to a real radar effect mechanism dataset for causality modeling and parameter optimization.

Conclusions:

• The BLMKM algorithm offers an efficient and accurate solution for Bayesian Network structure learning, particularly for high-dimensional and sparse datasets.
• The method has practical applications in establishing causality models, predicting outcomes, and guiding optimization in complex systems.
• BLMKM demonstrates significant potential for real-world data analysis and model development.