Research And Application Of Customer Segmentation Based On Ant Colony Clustering Algorithm

Ant colony algorithm has been successfully applied to solve a variety ofcomplex optimization problems because of its advantage of distributed computing,information feedback and heuristic search, however, the ant colony algorithm alsohas shortcomings as it has always been easy to fall into local optimum. Clusteranalysis is one of the important topics in data mining, especially its non-supervisedlearning mechanism, which do not need to specify the number of clusters. Studieshave shown that the combination of the ant colony algorithm and cluster analysis canbe a good solution to avoid the ant colony algorithm falling into local optimum,application of this could get a significant effect.This paper has researched the basic principles and idea of ant colony algorithm,described an example of its practical application, the traveling salesman problem,summarized each type of clustering algorithm, and highlighted the web-basedseveral clustering algorithm of which the input parameters affect more. Types of antcolony clustering algorithm has been studied, the ant heap-based clusteringalgorithm has following problems: complex parameter settings, moving randomnesscaused by the slow pace of convergence. Based on the study of standard ant colonyclustering algorithm, an improved adaptive ant colony clustering algorithm isproposed, the algorithm simplifies the setting of parameters, and introduces anadaptive strategy function, sets the similarity threshold which dynamically adjuststhe state of ants’ motion, lowers the mobile randomness, and finally applies theoriginal as the improved algorithm on processing the same data set, experimentalresults show that the improved one is smaller on the number of iterations, faster onconvergence and more efficient.Study found that the customer segmentation, which plays an important role inthe bank customer relationship management. Selecting the data of individualdeposits in a bank, which has been removed the name, gender, types of transactions,such as properties, in accordance with customer deposits, loans and monthly salary,the customers is divided into five categories, with the improved algorithm and K-means clustering algorithm to identify these data, the clustering results show thatthe improved algorithm in terms of recognition accuracy is better than K-meansclustering algorithm.