Big Data-driven Intelligent Analysis Of Targeted Poverty Alleviation

Targeted poverty alleviation is a worldwide problem as well as the important component of social governance.At present,we gain a lot of achievements about the theoretical research on targeted poverty alleviation from the perspective of sociology in China.At the same time,some applications which follow mode named “Internet + precision poverty alleviation” have been developed and already begun to take effect.However,it is rare to use internet technology to explore internal mechanism of poverty alleviation.Analyzing the principle of poverty and poverty alleviation quantitatively can not only improve the efficiency of aid effort,but also being a reference for studying the development of society and formulating policies.And it also provides a new direction for the job about international poverty alleviation.In order to improve the accuracy and efficiency of resource allocation and quantitatively analysis the potential rules in poverty alleviation work,we conducts relevant research on intelligent analysis of targeted poverty alleviation driven by poverty alleviation data in this paper.The main contents of the research contains that predict the time of getting rid of poverty and infer the rules for granting aid measures.For improving the suitability of the group for the assistance program and the accuracy of allocation about poverty alleviation resources,we propose a solution to infer the rules of aid measures using rule learning in machine learning,and select the most classic algorithm named Repeated Incremental Pruning to Produce Error Reduction(RIPPER)in rule learning to build the model.We try to infer the granting rules for the financial aid loan in education funding policies.In the first part of experiments,we use RIPPER and C4.5,PART decision tree to do the control experiment where generate three rulesets named R1,R2 and R3 respectively.The experimental results show that the coverage accuracy of R1 is 0.11 and 0.16 higher than that of R2 and R3 respectively.And on the other hand,the maximum ratio of recall for two types rules from R1 is 1.078,which is lower than 1.25 of R2 and 1.34 of R3.In the second part of the experiments,we keep the amount of every type equal which aims to contrast with the first part of experiment.In the second part of experients,the result shows that the maximum ratios of recall for two types of rules from R1,R2 or R3 is lower than that of first part of the experiment.The above results show that the RIPPER is stronger than the C4.5 and PART decision trees in terms of the coverage accuracy of ruleset about granting aid measures and the adaptability to unbalanced data.And at the same time,it is proved that the solution using RIPPER to infer rules for granting aid measures proposed in this paper is feasible and effective.For the lack of effective analytical models,we propose a prediction model for predicting poverty alleviation time based on FOA-BPNN,which aims to quantitatively analyze the effect of poverty alleviation and precisely predict the date of geting rid of poverty.In order to overcome the shortage of BP neural network model which may fall into the local minimum defect,the fruit fly optimization algorithm(FOA)is introduced.FOA use the error of training on BP neural network as fitness value to find the optimal BP neural network parameters and improve the accuracy of predicting.On the other hand,due to the fixed search radius of the standard fruit fly optimization algorithm,it may lead to weak the local optimization performance of FOA.DSFOA model using a dynamic step adjusting strategy is proposed which introduces the shift factor and population density.And then we combine BPNN with DSFOA to improve the prediction accuracy of the model.Based on the 50,000 samples of poverty alleviation in a poverty-stricken area in Hubei Province,experiment shows that the proposed DSFOA-BPNN model has improved prediction accuracy by 0.18 and 0.08 respectively compared with the BPNN and FOA-BPNN models.And At the same time,the results of incremental experiment improves that the proposed DSFOA-BPNN model is more adaptive in prediction accuracy.