Research On Location Selection Method Of Bank Outlets Based On Multi-source Data

With the full opening of China's financial industry to foreign-funded enterprises,China's financial industry is becoming increasingly fierce competition.As one of the important links of bank operation,the location of bank outlets has become an urgent problem for banks and other financial institutions to solve.And how choice the scientific and reasonable location is particularly important.Bank outlets needs to consider the factors is very complex when choices the locations.The factors are related,influenced deeply,and it is difficult to accurately described in mathematical model,and the relationship of the results and factors is nonlinear,and the traditional location method has certain drawbacks in the forecast,it is difficult to really provide theory support for the policy makers.In recent years,with the development of big data and machine learning technology,a lot of mapping relationships between input and output models can be learned without establishing complex mathematical equations,which makes it widely used in location problem.Therefore,this thesis will provide scientific and reasonable location decision for bank site selection by constructing a bank site selection model through combination of multi-source data and machine learning algorithms.The main work is as follows:(1)Combining with the research status at home and abroad,anlyzed the principles and influencing factors of bank outlets location.And determine the method of multic-source data processing and basic features of location selection in the work of this thesis.(2)Uses Pearson correlation coefficient and sequential backward selection method to select the optimal feature subset from feature set.Aiming at the problem of small sample size of labeled data,collaborative training method is used to dividing feature sets into two feature subsets as two views,traning a strong generalization classifier for each view,set confidence threshold,mark some unlabeled data and expand training set.(3)Building an ensemble learning algorithm based on stacking multi-model fusion and applying it to bank outlets location model.The thesis constructs two-tier learners,first tier is multiple base classifiers using five basic features.The second tier takes the predictive probability of the output of the first-tier learner as the feature,optimizes the loss function and constructs a classifier based on logistic regression.In order to ensure better location selection scheme ranking is more advanced.Finally,takes the location of a bank in Chongqing as the research objective and validates the algorithm by synthesizing a variety of evaluation indicators.The experimental results show that the algorithm in the thesis is superior to other algorithms with high accuracy and certain application value.