Application Of Double-Chain Quantum Genetic Algorithm In Mining Classification Rules

Data classification is a systematic method to create classification model based on input data sets,and classification rules mining is also an important research direction of data classification.Due to the superior performance exhibited by the evolutionary algorithm,after the genetic algorithm is applied to classification rules mining,many evolutionary algorithms such as decision tree,neural network,naive bayes,swarm-based algorithm and rule-based classification approach are widely used in the classification rules mining field.On the basis of reviewing and analyzing of the use of domestic and overseas evolutionary algorithms to solve classification rules mining problems,this paper analyzes the problems of using intelligent optimization algorithms to carry out classification rule mining,which is easy to fall into local optimal solutions,poor classification accuracy and robustness,then we present the Classification Rule Mining Based on Double Chains Quantum Genetic Optimization(DCQGA-CRM)and verify the effectiveness and practicability of the proposed algorithm in the field of classification rules mining through the performance analysis of the algorithm.The research content of this article mainly includes the following aspects:(1)The principle,updating strategy and flow chart of genetic algorithm,quantum genetic algorithm and double-chain quantum genetic algorithm are summarized,and then we use the three algorithms respectively to solve the problem of simple unary function optimization and multi-variate function optimization problem.The double-chain quantum genetic algorithm is used to solve the problem based on the evolutionary process analysis which has advantages compared to the other two algorithms.(2)Aimed at the deficiencies of intelligent optimization algorithms for mining classification rules,the Classification Rule Mining Based on Double Chains Quantum Genetic Optimization is proposed.The classification rule set is formed and implemented to achieve classification process from the quantum real coding,solution space transformation,quantum rotation gate updating and quantum variation.The classification process was compared with the Michigan algorithm,Pittsburgh algorithm,C4.5 algorithm and BP neural network to verify the effectiveness of the proposed algorithm in classification accuracy assessment experiment.(3)The concepts of class noise and feature noise are proposed.By adding different horizontal levels of noise to the feature attributes and class attributes of training data-sets,and then we can observe the variation trends of the prediction accuracy of the proposed algorithm compares with Michigan algorithm,C4.5 algorithm and BP neural network.At the same time,the concept of relative loss of accuracy(RLA)is introduced,we can analyze the classification rule mining algorithm based on DCQGA-CRM has better robustness and tolerance to noise under disturbance noise conditions compared with other proposed classification algorithms by calculating RLA.(4)In order to verify the proposed algorithm has obvious classification rule mining performance compared with other classification algorithms mentioned in this paper,non-parametric tests-Wilcoxon signed rank test method is introduced into the classification accuracy evaluation experiment and classification robustness evaluation experiment.The significance test shows that the performance of the proposed DCQGA-CRM algorithm is significantly improved compared with the contrast algorithm.