Increasing the explainability and success in classification: many-objective classification rule mining based on chaos...

Area of Science:

• Machine Learning
• Data Mining
• Computational Intelligence

Background:

• Classification rule mining is crucial for decision-making but often struggles to balance explainability with performance metrics.
• Existing metaheuristic methods offer potential but rarely optimize multiple objectives alongside interpretability.
• No prior method has simultaneously optimized more than three objectives while enhancing explainability for classification tasks.

Purpose of the Study:

• To propose a novel metaheuristic many-objective optimization-based rule extraction approach for classification.
• To introduce a chaotic SPEA2 algorithm for simultaneous optimization of four success metrics and automatic rule extraction.
• To enhance the explainability and interpretability of classification models.

Main Methods:

• Treating datasets as search spaces and metaheuristics as many-objective rule discovery strategies.
• Integrating chaos theory into the optimization method for performance enhancement.
• Utilizing a chaotic random search mechanism to mitigate issues like correlation and poor uniformity in candidate solutions.

Main Results:

• The proposed chaotic rule-based SPEA2 algorithm successfully optimizes four distinct success metrics simultaneously.
• Demonstrated ability to perform automatic rule extraction, enhancing model interpretability.
• Outperformed classical machine learning methods on three distinct datasets, showing improved efficacy.

Conclusions:

• The novel metaheuristic approach offers a significant advancement in classification rule mining.
• Simultaneous optimization of multiple objectives with enhanced explainability is achievable.
• The chaotic SPEA2 algorithm provides a scalable and interpretable solution for complex classification problems.