Design And Implementation Of Targeted Dynamic Identification System For Poor Households Based On Ensemble Learning

By the end of 2020,the targeted poverty alleviation work was successfully completed,but the elimination of absolute poverty and regional overall poverty does not mean the disappearance of poverty and the end of poverty alleviation work.The Central Committee of the Communist Party of China has repeatedly emphasized that it is necessary to improve the dynamic monitoring and assistance mechanism for preventing return to poverty,and implement normalized monitoring of the population vulnerable to return to poverty.At present,the research on the dynamic monitoring of poverty-returning mostly rests on the policy at macro level,while the operational research on the identification of poverty-returning at micro level is still lacking.The accumulated data of large-scale targeted poverty alleviation are not fully used.With the success of poverty alleviation,preventing return to poverty in the post-2020 era has become the focus of follow-up work.In view of the above problems,this paper sorts out the theory and experience of anti-poverty at home and abroad,especially the content of poverty alleviation by means of information technology.Combined with the results of China's targeted poverty alleviation work and targeted poverty alleviation system construction,this paper makes an empirical study based on the data of poor households in Fuping County.By comparing the results of traditional machine learning classification algorithm and ensemble learning algorithm,this paper constructs a poverty population identification model based on ensemble learning algorithm and classifies the poor households as population out of poverty,poverty and returning to poverty.The results show that the optimized XGBoost algorithm model achieves the best results,and the identification accuracy of the three categories of people who have been out of poverty,poverty and returning to poverty is 97.43%,92.44% and97.04% respectively,and the overall accuracy is 96.81%,which can better identify the poverty category of the poor population.At the same time,the ranking of the importance of features in the classification of the model can reveal the importance of poverty causing features of poor households in Fuping County,and help the poverty alleviation countermeasures to be implemented more targeted.Through analyzing the current work process of helping personnel,we have made it clear that,in order to do a good job in the dynamic monitoring of poverty-returning,we need to achieve the dynamic updating of data,dynamic feedback of results and dynamic adjustment of countermeasures.In addition,flask framework is used to combine the call of algorithm model with the targeted poverty alleviation platform to develop a precision dynamic identification system for poor households,which can satisfy the demands of real-time identification of poverty status of the helping objects,and supplement the deficiency of the combination of most current research and practical application scenarios.It provides practical case support for the dynamic monitoring and early warning of poverty-returning in the "post poverty era".