Multiage Evolutionary Algorithm and Its Application in Data Mining

Mining association rule form a large database is a time-consuming task for evolutionary algorithms when the main memory is limited and the given dataset is very large. In this paper, we propose a novel multiage evolutionary algorithm (MAEA) to mine the rules more efficiently, accurately and reliably. Inspired by the natural evolutionary process, in MAEA, individuals have a flexible life span during which they can grow and reproduce independently. Different from conventional GAs, in which the whole dataset is checked one time for each individual, the MAEA algorithm calculates component of fitness function of different individuals simultaneously when a record was read in, and the whole fitness of individuals in a population can be got when the whole records are checked one time. The experimental results showed that the MAEA algorithm is superior to the traditional genetic algorithm (GA) in searching for the optimal solutions when the number of the records is very large and the internal memory is limited.