Improving rule-based classification using Harmony Search

Area of Science:

• Data Mining and Machine Learning
• Artificial Intelligence
• Pattern Recognition

Background:

• Rule-based classifiers integrate classification and associative rule mining for enhanced decision-making.
• Existing methods often employ greedy techniques, leading to suboptimal rule selection and high processing overhead.
• Traditional algorithms struggle to effectively explore the search space for the best rule subsets.

Purpose of the Study:

• To develop an advanced rule-based classifier addressing limitations of existing methods.
• To improve classification accuracy and reduce processing overhead in rule mining.
• To effectively select a globally optimal subset of classification rules.

Main Methods:

• Integration of the Apriori algorithm, Harmony Search, and Classification-Based Association Rules (CBA).
• Application of a modified Apriori algorithm with multiple minimum supports for class-specific rule extraction.
• Utilization of binary Harmony Search for efficient selection of the best rule subset for classification models.

Main Results:

• The proposed method was evaluated on seventeen benchmark datasets.
• Statistical analysis demonstrated superior performance compared to traditional association rule classification algorithms.
• The approach effectively overcomes limitations related to redundant rules and high computational costs.

Conclusions:

• The hybrid approach combining Apriori, Harmony Search, and CBA significantly enhances rule-based classification.
• The method provides a more effective strategy for exploiting the search space to identify optimal classification rules.
• This novel technique offers a promising advancement for applications requiring accurate and efficient rule-based classification.