An effective feature learning approach using genetic programming for crab age classification

Reliable age estimation plays an important role in managing populations of marine organisms. The extraction and analysis of eyestalks and gastric mill ossicles for determining the age of crabs are difficult and extremely time consuming. In this paper, we propose a novel Genetic Programming (GP) approach to learning high-level features from easily accessible features of crabs, such as length, weight, and sex, for crab age classification. We develop a new representation of GP to extend the width and depth of GP trees, so as to automatically generate a flexible number of high-level features without extensive domain knowledge. With the high-level features and easily accessible features, the new GP approach is subsequently wrapped with classifiers, e.g., Support Vector Machine (SVM), to effectively classify the crab age. The performance of the proposed GP approach is compared with five mainstream machine learning classification algorithms. Experiments show that the high-level features learned by GP improve the classification accuracy of crab age classification. Moreover, the learned features have good interpretability.