Artificial Immune System And Its Application In Data Processing

The artificial immune system is a new artificial intelligence model inspired bybiological immune system which gives researchers many inspirations. Research ofAIS is later than the neural network model, therefore there are still many points tostudy. This paper proposed a algorithm named Parallel Clonal Selection AlgorithmFor Multi-class Classification(multi_CSA) and proved multi_CSA’s convergence intheory. Since multi_CSA is a supervised classification algorithm that its performancewill be affected by sample data. In order to improve the algorithm in efficiency andmake full use of priori knowledge of train sample, paper then proposed a algorithmnamed Data Purification and Clustering Algorithm Based on Artificial ImmuneNetwork(aiCA) and analyzed the convergence of aiCA. Machine learning datasetsand remote-sensing images were used for multi_CSA to classify data and aiCA todata purification and clustering in application. The detailed contents are as follows:(1) multi_CSA were proposed in this paper to resolve the problem that traditionalCSA could only provide supervised learning for a certain type of sample data whichmay result in lower classification efficiency and accuracy. Algorithm could obtainoptimal clustering center of the sample data at the same time through new fitnesscalculation and selection policy.(2)This paper used UCI data and mangrove multispectral TM images formulti_CSA classification experiments respectively. Experiment obtained the overallclassification accuracy of91%and Kappa coefficient of0.89, UCI data also got agood result.(3)Two random convergence measures(complete convergence and meanconvergence) were used to analyze the convergence of multi_CSA. Paper woulddemonstrate that multi_CSA satisfied the sufficient condition for convergence.Experiment also was performed to validate the results later. Conclusion was got thatmulti_CSA would converge to the global optimum with probability1inexperimentally and theoretically. (4) aiCA was proposed through using the prior knowledge of train sample. aiCAcould better reflect the distribution of training data which resulted in the improvementof clustering and generalization ability.(5)First，this paper used aiCA in data purification and compression to solve theproblem that multi_CSA influenced by the sample data and improve the classificationaccuracy、efficiency of the algorithm. Second, aiCA was used in image clustering toavoid issue that multi_CSA need sample data.(6)This paper proved the convergence of aiCA using Bayesian conditionalprobability.(7)A prototype was built to study the results visually and easily. Prototype includedsome simple processing of the sample as well as algorithms run.In this paper, we studied the clonal selection algorithm and artificial immunenetwork algorithm in theory, then proposed two improved algorithm and proved itsconvergence respectively. Remote sensing image and machine learning dataclassification were used to validate the two algorithm. All experiment’s resultsshowed that the contents of study in this paper were feasible and effective in practiceand theory.