Feature selection in medical text classification based on immune algorithm

In text classification, effective feature selection is essential to make the learning task more efficient and accurate. This paper proposes a new feature selection algorithm based on Immune Clonal Selection Algorithm (ICSA) for medical text classification according to the characteristics of medical document. It considers that the affinity based on Jeffries-Matusita distance and the clone operator can sure to gain the property of rapid convergence to global optimum, which speeds up the searching of the most suitable feature subset among a huge number of possible feature combinations. The experimental results show that the classification accuracy in medical document is improved effectively and characteristic dimension is reduced a lot. Compared with BP neural network(BP) and genetic algorithm (GA), the proposed method can find better feature subset for classification in the limited number of evolutionary generations.