Discovering Classification Rules by Real Coded MOGA

Classification problem is one of the wellstudied problems in machine learning. By supervised way we can solve it. First step is the generation of IF-THEN rules at the learning phase from records where class is known. Second step is to use those rules to classify records where class is not known. In this article we are using real coded multi objective genetic algorithms (MOGAs) for generating a set of optimized classification rules. Real-valued attribute ranges are encoded with real-valued genes and we are presenting a new genetic algorithm operator. Rules set are optimized from two objectives – confidence and coverage within the scope of our present scheme. All rules are non-dominated and lie on the pareto-optimal front. Results demonstrate effectiveness of classification rule generation by real coded multi objective genetic algorithm (CRGRCMOGA) by applying it on 8-benchmark data set from UCI machine learning database.