Classification results of coronary heart disease database by using the clonal selection method with receptor editing

Abstract—The clonal selection principle is used to explain thebasic features of an adaptive immune response to a antigenicstimulus. It established the idea that only those cells thatrecognize the antigens are selected to proliferate and differentiate.This paper explains a computational implementation of theclonal selection principle that explicitly takes into account theaffinity maturation of the immune response. The clonal selectionalgorithm by incorporating receptor editing method, RECSA, hasbeen proposed by Gao. This paper tries to classify the medicaldatabase of Coronary Heart Disease databases and reports thecomputational results for 4 kinds of training datasets. I. I NTRODUCTION In a few decades, the area of artificial immune system (AIS)has been an ever-increasing interested in not only theoret-ical works but applications in pattern recognition, networksecurity, and optimizations[1],[2],[3],[4],[5],[6],[7]. AIS usesideas gleaned from immunology in order to develop adaptivesystems capable of performing a wide range of tasks in variousresearch areas. Especially, Gao focuses the Clonal SelectionAlgorithm(CSA) which is known to one of the famous immunealgorithms and proposed the effective method for solvingTraveling Salesman Problem(TSP)[8] which is known to beNP-complete.The biological basis of the Clonal Selection Theory wasproposed by Burent[13],[14] in 1959. The theory interpretsthe response of lymphocytes in the face of antigenic stimulus.Only the immune cells with high affinity are selected toproliferate, while those low affinity cells must be efficientlydeleted or become anergic. The hypermutation is allowedto improve the affinity of the selected cells to the selectiveantigens. Receptor Editing as a mechanism of immune celltolerance is reported[9],[10],[11].Gao indicated the complementary roles of somatic hyper-mutation (HM) and receptor editing (RE) and presented anovel clonal selection algorithm called RECSA model byincorporating the Receptor Editing method[8]. In [8], theydiscussed the relationships between HM and RE throughutilizing them to solve the TSPs. Because a valid tour in TSPis represented by a permutation of N cities, the number ofstates to feasible tours is (N −1)!. [1] solves the problems forfinding an optimal set in the search space consisted of the setof (N − 1)! valid tours in TSPs.A medical database named Coronary Heart DiseaseDatabase (CHD DB) has been prepared in the data miningcontest. The database makes it possible to assess the effective-ness of data mining method in medical data. The CHD DB isbased on actual measurements of the Framingham Heart Study- one of the most famous prospective studies of cardiovasculardisease. It includes more than 10,000 records related to thedevelopment of coronary heart disease (CHD). The datasetshave been proved enough valid by statistical analyses[12].This paper challenges to classify the CHD DB by usingRECSA model. The classification of medical database differsfrom the TSP, because medical information such as resultsof biochemical tests and chief complaint is often ambiguous.Therefore, we cannot clearly distinguish the difference be-tween normal and pathological values. Biochemical test valuescannot be precisely evaluated by using crisp sets. In thispaper, we consider that the database has some relation betweeninputs and output signals. That is, an output is summed upall the inputs modified by their respective weights and iscompared with the corresponding teach signal. We report thecomputational classification results of the databases.The remainder of this paper is organized as follows. In thenext serction, the clonal selection theory will be explainedbriefly. Section 3 will explain RECSA model proposed by [8].The CHD DB will be described in Section 4. Experimentalresults for classification of the CHD DB will be reported inSection 5. In Section 6, we give some discussions to concludethis paper.II. T