Predicting the Tide with Genetic Programming and Semantic-based Crossovers
Nguyen Quang Uy, Michael O’Neill, Nguyễn Xuân Hoài · 2010
This paper proposes an improvement of a recently proposed semantic-based crossover, Semantic Similarity-based Crossover (SSC). The new crossover, called the Most Semantic Similarity-based Crossover (MSSC), is tested with Genetic Programming (GP) on a real world problem, as in predicting the tide in Venice Lagoon, Italy. The results are compared with GP using Standard Crossover (SC) and GP using validation sets. The comparative results show that while using validation sets give only limited effect, using semantic-based crossovers, especially MSSC, remarkably improve the ability of GP to predict time series for the tested problem. Further analysis on GP code bloat helps to explain the reason behind this superiority of MSSC.