Using Ant Programming Guided by Grammar for Building Rule-Based Classifiers.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Extracting comprehensible knowledge is a significant challenge across many scientific and industrial domains.
• Existing ant programming (AP) algorithms are not specifically designed for mining easily understandable classification rules.

Purpose of the Study:

• To present a novel ant programming (AP) framework, termed grammar-based ant programming (GBAP), for extracting human-comprehensible classification rules.
• To enhance decision support systems by providing interpretable classification models.

Main Methods:

• Developed the grammar-based ant programming (GBAP) algorithm, guided by a context-free grammar to ensure valid rule generation.
• Introduced two complementary heuristic functions to compute movement probabilities, improving upon traditional single-heuristic methods.
• Employed a niching approach for selecting rule consequents and constructing the final classifier.

Main Results:

• The GBAP algorithm successfully generates classification rules that are easily understood by humans.
• Comparative experiments on 18 diverse datasets demonstrate that GBAP achieves accuracy comparable to or better than existing classification techniques.
• The generated rules effectively support expert-domain decision-making.

Conclusions:

• Grammar-based ant programming (GBAP) is an effective method for extracting interpretable classification rules.
• GBAP offers a valuable tool for knowledge discovery and decision support in data mining.
• The dual-heuristic and niching strategies contribute to the performance and interpretability of the generated models.