A Hybrid Artificial Immune Algorithm for Feature Selection of Ovarian Cancer Data

The paper presents a novel hybrid immune algorithm (HAIA) and applies it to the feature selection of SELDI-TOF MS ovarian cancer data to select the best subset of features for medical decision. Based on an immune metaphor, the algorithm is integrated with clonal selection, bone marrow model, ant colony optimization, and K nearest neighbor classifier. Clonal selection presents a local search and suppression mechanism which is different from traditional search methods, showing an effective performance on this problem. Bone marrow model creates macrophages for antigen presentation, and ant colony optimization is applied to training vaccines without any experience. The selected subsets are tested by K nearest neighbor classifier and evaluated with classification accuracies. The feature selection and classification of ovarian cancer datasets are done with 5-fold cross validation method. The experiment results of the classification accuracy, number of resources and classification time with this new algorithm are compared with some reported methods and traditional algorithms which are applied to related problems. HAIA, reaching a classification accuracy of 93.2% for ovarian cancer, obtained the highest classification accuracy among those algorithms. Besides, HAIA gains another advantage over those algorithms by reducing the cardinality of the selected subset to 25 from the whole 84 features. For these reasons, HAIA manifests that it could be used as a much effective way for medical problems.