Bank Telemarketing Customer Classification Based On Multi-Core Extreme Learning Machine

With the advent of the Big Data era,banks are finding valuable customers by traditional telemarketing in an ever-increasing amount of customer information,which is like looking for a needle in a haystack and requires a lot of time and human resources.Therefore,how to accurately classify the huge number of customers has become the key to bank telemarketing.Machine learning methods are widely used due to their powerful data mining capabilities,and have shown significant performance advantages in classification problems.In view of this,this thesis adopts the improved machine learning method to classify and predict bank telemarketing customer data in an attempt to improve the hit rate of bank telemarketing.To further improve the accuracy of bank telemarketing predictions,this thesis proposes a multi-core extreme learning machine classification method(Gt SSA-MKELM-KA)based on an improved sparrow search algorithm optimization for accurate classification of bank telemarketing customers..Firstly,the Boruta algorithm-based method for selecting important features for customers is proposed,which significantly reduces the computational complexity of the model by reducing the number of feature dimensions.Secondly,the ADASYN adaptive integrated sampling technique was used to imbalance the data set so that the number of positive and negative category samples was balanced.Finally,this thesis improves the extreme learning machine based on the multi-kernel function,and constructs a multi-core extreme learning machine model(MKELM)based on multiple Gaussian functions.Since the performance of the multi-kernel function is easily affected by parameters,this thesis uses the sparrow search algorithm to parameterize the multi-kernel function.In order to optimize the sparrow search algorithm,an improved sparrow search algorithm(Gt SSA)based on the good point set theory and adaptive t distribution variation is proposed.In addition,in order to reduce the computational complexity of the model,a Gt SSA fitness function definition based on the Kernel Target Metric(KA)is proposed,and combined with the previous improved classification method,the Gt SSA-MKELM-KA classification model used in this thesis is finally constructed.For the customer classification results,this thesis introduces three clustering ideas,analyzes the customer classification results predicted by the classification model,and then divides the customer groups into more fine-grained categories.In this thesis,experiments are carried out in combination with the real data of telemarketing of banking institutions.Comparing the improved method proposed in this thesis with the original method and other popular machine learning methods,the experimental results show that the model in this thesis has a better classification effect on bank customers in terms of AUC,Acc and other indicators.Based on the three-branch clustering model,the customer classification results are fine-grained divided.Finally,customers are divided into ordering customers,non-ordering customers and potential customers,and TSNE visualization is carried out.Through the comparison of customer classification before and after fine-grained division,it is found that the division method proposed in this thesis is more reasonable.To sum up,the customer classification model proposed in this thesis has a good performance in the actual bank telemarketing scenario,which can provide assistance for bank telemarketers to carry out telemarketing activities more accurately.