Study On Data Classification Learning Model Based On Swarm Intelligent Algorithm

In the era of massive data, people’s urgent need for techniques that can automatically and intelligently transform data into useful knowledge promotes the rapid development of data mining techniques. Data classification is one of the most important tasks in data mining, it can be used to discover the model that describes important data class as well as predicting the trend. Data classification has been made extensive research in artificial intelligence, network finance,pattern recognition, machine learning and other fields, and also produced a large number of classification modeling algorithms. Although data classification has made some breakthroughs theoretically and technically, it still faces many challenges, mainly including the accuracy and efficiency of classification algorithms, the comprehensibility of classification rules. The traditional classification algorithms are facing great challenges in efficiency, comprehensibility and scalability.Because of mining classification rules is just an optimized process to find out the best rules set from training databese, so swarm intelligence algorithm is used to construct classification model by many researchers and achieved some results.Swarm intelligence algorithm simulates a population of nature in which individuals cooperate to solve a problem. This kind of algorithm has potential concurrency and strong robustness, andit doesn’t depend on specific problems. What’s more, it can solve pretty complicated problems through cooperation between individuals. In recent years, applying swarm intelligence algorithm to classification modeling attracts many researchers and is becoming a focus of academic research.This paper focuses on constructing classification learning model of ant colony algorithm and gene expression programming algorithm, which are representational in swarm intelligent optimization algorithm, andproposes a new ant colony classification modeling algorithm Ant-Miner PAE and a new gene expression programming algorithm IGEP, then validates the effectiveness by several experiments. The main research work of this paper is as follows:(1)After introducing the definition, principle and implementation method of data classification and analyzing the shortcomings of the common classification modeling algorithm, this paper researches on the ideological source of ant colony algorithm, working methods and key steps, and further studies the traditional ant classification algorithm Ant-Miner and its development history, and then proposes a new ant colony algorithm based on pheromone attraction and exclusion to avoid premature and local extremum, designing new equations of computing pheromone and transitional probability, adjusting the order of iterations and adopting global pheromone updating and local pheromone updating methods, in which pheromone contains attractive part and excusive part in the searching procedure of ants, this method can balance the relationship between exploration and development, making the ants explore initially and develop in the late period.(2) In order to validate the effectiveness of the improved algorithm, this paper applied the improved algorithm to classification modeling problems and takes the influences between rules into consideration. This paper proposes a new ant classification modeling algorithm Ant-Miner PAE, conducts experiments using 12 public datasets of UCI, and compares the experiment results with other classification algorithm such as CN2, C4.5rules, PSO/ACO2, Ant-Miner and c Ant-Miner PB.The experimental result shows that the improved algorithm has advantages in prediction accuracy and rule simplicity.(3) After researching on the ideas of gene expression programming as well as its shortcomings, this paper proposes an improved GEP algorithm with“gene extraction”, “geneturnover” and“gene dynamic adjustment” in order to overcome its shortcomings such as being easily affected by noise and causing premature convergence when modeling. The improved algorithm can not only ensure the diversity of population during the whole evolution process, but also improve the gene utilization rate, making the algorithm have higher convergence rate and accuracy as well as avoid premature.(4) In order to prove the advantages of the improved GEP algorithm, this paper applies the algorithm to model and predicts the vegetable price from 11 May 2015 to 5 August 2015 in China. Combined with time series method, the training data were analyzed and evolved, this paper built a mathematical model to implement the simulation and predication on the vegetable price. The experiment also implements linear regression, parabolic regression, simulation and prediction of basic GEP algorithm.Multiple experiments and comparison with other algorithms prove that the improved GEP algorithm has faster convergence speed and higher accuracy.(5) Based on the improved GEP algorithm, this paper proposes the concepts of balanced probability selection and super function that make the terminators and operators are selected in the same probability, the operators in different number can be also selected in the same probability and the characters and functions of the mutation operator have equal opportunity to choose, which ensures the diversity of population. The setting of super function can make it jump out of the local optima and search better solution. Then the new GEP algorithm is used to constructing classification learning model aiming at the binary class classification problem such as breast cancer, balloon, credit card and three-class problem such as wine recognition and iris classification. Utilizing the new GEP algorithm to conduct experiments and comparison with basic GEP algorithm and other algorithms shows that the new GEP algorithm has better accuracy in classification prediction.