New Classification Method for Independent Data Sources Using Pawlak Conflict Model and Decision Trees

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Managing independent datasets in local decision tables often involves building individual classifiers.
• Existing literature focuses on combining classifier predictions, with limited research on inter-table cooperation and coalition formation.

Purpose of the Study:

• To propose a new method for creating coalitions of local tables to enhance classification quality and reduce computational complexity.
• To investigate the impact of table cooperation on learning generalized concepts and improving new object classification.

Main Methods:

• A novel method for generating coalitions of local tables using attribute value characteristics and Pawlak conflict analysis.
• Building classification and regression trees with Gini index on aggregated tables for each coalition.
• Aggregating decisions from coalition classifiers using majority voting in a hierarchical structure.

Main Results:

• The proposed coalition-based approach significantly improved classification accuracy (0.847 vs. 0.812) compared to non-cooperative methods.
• The algorithm implementing the coalition approach was up to 21 times faster than the non-cooperative method.
• Statistical analysis (Wilcoxon test, p=0.005) confirmed the significance of accuracy improvements.

Conclusions:

• Forming coalitions of local decision tables enhances classification quality and efficiency.
• The proposed method offers a generalized approach to concept learning and improved classification performance.
• This research highlights the benefits of inter-table cooperation in data mining and machine learning tasks.