Decision Tree Integration Using Dynamic Regions of Competence

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Base model integration is crucial for Multiple Classifier Systems (MCS).
• Existing methods like Random Forest use majority voting for fusing base classifiers.
• There's a need for advanced integration techniques in MCS.

Purpose of the Study:

• To propose a new MCS algorithm for base model integration.
• To enhance classification performance by introducing a novel weighting strategy.
• To validate the proposed method against existing MCS approaches.

Main Methods:

• The proposed algorithm partitions feature space based on decision tree rules.
• Centroids of partitioned subspaces are calculated.
• Weights for base classifiers are determined using these centroids for a weighted majority voting rule.

Main Results:

• The proposed MCS algorithm demonstrated effectiveness on open-source benchmarking datasets.
• Experimental results show superior performance compared to other MCS approaches.
• Micro and macro-average classification measures were used to analyze performance.

Conclusions:

• The novel feature space partitioning and centroid-based weighting method improves MCS integration.
• The proposed algorithm offers an effective alternative for building robust Multiple Classifier Systems.
• The study highlights the potential of this approach for various classification tasks.