| China’s crude divorce rate,which has risen to 57 percent,is one of the hottest social topics,with the national civil affairs administration announcing that 3 million couples will get divorced in 2019.At the same time,the current neural network artificial intelligence development,which provides technical support for forecasting,classification.Therefore,the purpose of this paper is to use machine learning,statistical knowledge and technology to establish a divorce prediction evaluation model and analyze the causes of divorce.In this paper,a predictive divorce model is established by using machine learning method for feature-based samples,and the main causes of divorce are analyzed and corresponding Suggestions are provided according to the machine learning algorithm.Firstly,from the research background,research significance and research status at home and abroad,this paper fully explains the feasibility and necessity of the research,and provides the premise for the following research.Secondly,the original samples were processed by data pretreatment method: first,the chi-square test of independent variables and dependent variables was used in statistics to test the independence of each independent variable and dependent variable.It was found that there was no independent relationship between all independent variables and dependent variables in the data set,that is,each attribute was related to the dependent variable.Second,to increase the use of Spearman correlation coefficient analysis of the correlation between each independent variable,according to the research conclusion found features 6 and 2 there is no correlation between(that is,both sides of husband and wife can be soul mates and their differences between itself and no statistically significant correlation between),features 6 and there is no correlation between 45(that is,the couple would communicate and they have no significant difference between the relevant relations),and 10 of variables between highly relevant,delete one of the variables can be arbitrary.Delete data sets 10 characteristics,therefore,are characteristics of 4(need)empathy were arguing with each other,and features 11(rapport),17(common),features(know each other’s anxiety),26,29(get to know each other),features 35(insult each other),22(know how to care for a spouse),features 10(common goals),36(when discussing humiliated),characteristics of 38(hate open approaches spouse).Then,after a summary of this research take the method principle and concept,using data preprocessing,after using the decision tree and random forest,the BP neural network,GBDT four machine learning algorithms fitting prediction model,by constantly adjust the model parameters,find out the optimal model and the accuracy and recall rate,precision and AUC value evaluation model.The empirical analysis found that GBDT predicted the divorce model with the best comprehensive performance--accuracy 0.9811321,recall 0.962963,difference accuracy 1.000000 and AUC value 0.9957.Other models also have corresponding advantages in other performance indexes,so it is necessary to choose the model according to the actual situation and make the correct marriage prediction behavior.Finally,according to the model of important features sorting,18 / view of marriage have a common characteristic,features 19(the role of the two sides have a common view in marriage),characteristics of 34(argument using hate expression),features 40(between the other did not happen,just talking about it),features 20(in terms of trust,the two sides have the same values),features 16(both sides in love compatibility),the characteristics of 39(sudden discussion between both sides),the characteristics of 9(likes to travel with your spouse)belongs to the important characteristic,also is the important cause of the divorce,according to the above conclusions,We can understand what the main causes of divorce are and put forward corresponding Suggestions for these important reasons in order to reduce the risk of divorce and further reduce the divorce rate... |
