A Swarm Optimization approach for clinical knowledge mining.

Area of Science:

• Data Mining
• Artificial Intelligence
• Bio-inspired Computing

Background:

• Rule-based classification is crucial for medical diagnosis and decision support systems.
• Optimizing rule sets using heuristic or meta-heuristic approaches improves classification efficiency.
• This study introduces Wind-driven Swarm Optimization (WSO), a novel biologically inspired meta-heuristic algorithm.

Purpose of the Study:

• To apply the Wind-driven Swarm Optimization (WSO) algorithm for optimizing rule bases in medical diagnosis.
• To evaluate the efficiency of WSO in generating accurate and concise rule sets for classification tasks.
• To compare WSO's performance against traditional methods like Particle Swarm Optimization (PSO) and decision tree algorithms.

Main Methods:

• WSO was employed to optimize rule sets extracted from decision trees.
• A new metric, Jval, was utilized to assess the efficiency of the rule-based classifier.
• The performance of WSO was analyzed across various decision tree extensions (RIPPER, PART, FURIA, Decision Tables) and compared with PSO.

Main Results:

• WSO demonstrated improved accuracy and conciseness compared to the traditional C4.5 algorithm on benchmark medical datasets.
• For liver disorders, WSO achieved 64.60% accuracy with 19 rules versus C4.5's 62.89% with 43 rules.
• For Heart disease, WSO achieved 77.8% accuracy with 34 rules versus C4.5's 68.64% with 98 rules. Normalized standard deviations for PSO and WSO accuracy were 0.5921 and 0.5846, respectively.

Conclusions:

• WSO effectively generates accurate and concise rule sets, outperforming traditional methods in medical data classification.
• While PSO offers similar results, WSO's novelty lies in its biological inspiration and customization for rule base optimization.
• Optimizing the trade-off between prediction accuracy and rule base size is key for effective clinical decision support systems.