Initial population composition methods for evolutionary outcome-accumulating rule discovery

Abstract

GNMiner has been proposed as an outcome accumulation and evolutionary knowledge discovery method. This method discovers and accumulates rules and evolves by utilizing the acquired information, but the time required to acquire initial outcomes accounts for a large proportion of the total time required. To improve the computational efficiency at this stage, we propose two approaches: one is to use prior information from the data of the rule discovery target, and the other is to use a replicated population stored at a certain generation in the evolutionary process. Simulations were performed in multiple settings for each approach to evaluate the evolutionary process and the impact on the diversity of the discovered rule set. The results showed a reduction in the time required to obtain initial outcomes, and provided insight into each approach.