Research On The Problem Of Accurate Loan Forecasting For Poverty Alleviation And Accurate Recommendation Of Poverty Alleviation Measures

Poverty is a complex and realistic social problem.The factors affecting the poverty of farmers are diverse.By analyzing the data characteristics of poor households,mining the key influences and driving factors from multiple influencing factors,and establishing some predictive models of poverty alleviation,it is of great significance for Targeted Poverty Alleviation.The data of poverty alleviation come from a wide range of sources(basic information of poor households,medical institutions,social security,education,etc.),are diverse in types and are updated at a fast speed,which effectively promotes the implementation of policies of Targeted Poverty Alleviation.At the same time,they also have produced massive amounts of data of poverty alleviation.How to dig out valuable information for the fight against poverty from these abundant data of poverty alleviation is one of the challenges facing targeted poverty alleviation work.Based on the two perspectives of objects and measures of poverty alleviation,this paper explores and analyzes the key data of Targeted Poverty Alleviation in Gansu province,and studies the following two aspects through machine learning.Based on the Back-Propagation Neural Network(BPNN)model is a prediction model of precision loan for poverty alleviation objects.The purpose of this model is to predict whether poor households can apply for the special loans for Targeted Poverty Alleviation,and provide financial guarantee for poor households with potential for industrial development but lack of funds.With the exponential growth of high-dimensional data of the targeted poverty alleviation,the traditional methods of training BPNN has slow convergence,low efficiency and easy to fall into local optimization in the process of training model for precision loan prediction.Although a large number of scholars have proposed training BPNN by distributing parallel,there are still the following problems:(1)how to find the global optimal solution from their respective slices;(2)how to avoid falling into local convergence in the process of model training.To solve the above problems,this paper proposes an evolutionary training method of BP neural network model of precision loan prediction using Spark.The method will filter the local weight matrix that generated during the training process to form the initial population,which will be evolved by genetic algorithm to enhance the ability of global convergence,and the individual with the highest fitness is selected as the initial weight matrix for the next iteration.In this way,not only the number of iterations for global optimization can be reduced,but also the local convergence can be effectively avoided on the high-dimensional data set of poverty alleviation.Experimental results show that this method improves the speed of convergence and accuracy of the model of precision loan prediction for poverty alleviation objects based on BPNN.The precision recommendation model of poverty alleviation measures is based on FCM(Fuzzy C-Means).The purpose of the precision recommendation help's measures is to increase the self-production capacity and create economic sources of poor households,so that they have "hematopoietic" ability to get rid of poverty and become rich.At present,many training methods of the FCM clustering model are faced with high time and space complexity when dealing with large data sets,which makes it difficult to complete the training of accurate recommendation model for poverty alleviation measures on a single computer,thus affecting the application in reality.In order to solve the above problems,this paper proposes a novel training method of precision recommendation model for poverty alleviation measures of FCM.Combining with the parallel computing framework of Spark,which is a fast and memory-based iteration cluster computing system,multiple data splits of poverty alleviation are trained in parallel to obtain several FCM recommendation models.The out of each model is a set of clustering centers and corresponding assistance measures.In the assistance measures of each cluster center,each measure has its own occupation ratio.The final result of the precision recommendation measure is a vote of multiple FCM recommendation models.Experiments show that the proposed training method of recommendation model for assistance measures has potential advantages in terms of time,accuracy,speedup and scalability.In summary,by integrating the large-scale data of poor households with multi-source heterogeneous data and combining analytical methods of machine learning,this paper builds a prediction model of precision loan for poverty alleviation objects based on BPNN and a precision recommendation model of assistance measures based on FCM.Through the multi-dimensional modeling and analysis of the large data of poverty alleviation,it provides reliable theoretical support for Targeted Poverty Alleviation from multiple aspects and has a reference significance for improving the accuracy and effectiveness of poverty alleviation decision-making.