A Feature-based Deep Neural Framework for Poverty Prediction

Abstract

Poverty eradication has been a challenge for humanity, especially when it comes to sustainability, if a family with a tendency to sink back into poverty can be detected, with necessary assistance provided in advance, such situation may be avoided. In order to detect them, we design a data-driven procedure to capture the relevance between poverty and related data, putting forward a deep neural multi-channel model to encode multi-type features, named Deep Poverty Forecast(DPF). In this paper, we also introduce a star graph to represent the relationship between family members and a data labeling method for supervised learning. Extensive experiments are conducted over five city datasets and our results show that our proposed framework has achieved much gain over previous methods. This solution applied at Data Platform of Poverty Reduction in Guangxi province has covered 84% of new families falling into poverty while the search space is reduced by over 90% according to the survey conducted by the local government. Our proposed framework can be easily applied to other family-related scenarios as well.