A hybrid selection algorithm for time series modeling

Research paper by Julie Yu-Chih Liu, Juo-Chiang Hsieh

Indexed on: 23 Feb '14Published on: 23 Feb '14Published in: Soft Computing


An evolutionary algorithm becomes trapped in local optima when a premature convergence occurs. Research has suggested maintaining population diversity to address this problem. However, traditional methods are excessively complex and time consuming. This study proposes a hybrid selection mechanism in which clonal and roulette wheel selections are alternated to maintain population diversity during evolution. The proposed method is based on a genetic programming technique known as gene expression programming (GEP). The prediction power and efficiency of the proposed method were compared with those of other GEP-based algorithms by using five time series benchmarks. The experimental results indicated that the proposed algorithm outperforms the other algorithms.