A Feature Construction Method That Combines Particle Swarm Optimization and Grammatical Evolution

The problem of data classification or fitting is widely applicable in a multitude scientific areas, and for this reason, number machine learning models have been developed. However, many cases, these present problems overfitting cannot generalize satisfactorily to unknown data. Furthermore, the features input do not contribute learning, there may even be hidden correlations between dataset. purpose proposed method significantly reduce regression errors through usage technique that utilizes particle swarm optimization grammatical evolution. This divided into two phases. In first phase, artificial are constructed using evolution, progress creation controlled by method. addition, new penalty factors limit generated range values make training more efficient. second phase technique, exploited transform original dataset, then any can applied performance was measured on some benchmark datasets from relevant literature. Also, tested against series used models. experiments performed showed significant improvement 30% average an greater 60% datasets.