Mining Classification Rule with Artificial Fish Swarm
Meifeng Zhang, Cheng Shao, Meijuan Li, Junman Sun · 2006
As a novel simulated evolutionary computation technique, artificial fish swarm algorithm (AFSA) shows many promising characters. This paper presents the use of AFSA as a new tool for data mining to discover classification rules from data, called AF-Miner. Mining classification rule task is formulized into an optimization problem. Furthermore, each potential if-then rule is encoded into a real-valued artificial fish (AF) that contains the upper and lower limits of all attributes in data sets. The simulation results show that AF-Miner can mine better classification rule, including rule set with higher predictive accuracy rate, better generalization ability and the smaller number of rules, simpler rule with fewer terms. And also show that the new approach has good performance for rule discovery on continuous data