Poverty Prediction Through Machine Learning

Poverty elimination stands as an inevitable process in human development, with predicting poverty being the first and one of the essential steps. The paper considers poverty as an outcome of multidimensional factors, and offers various practical models for such prediction using machine learning, none of which accounts for the whole, while some factors may outweigh others. Thereby, an integrated approach of prediction is needed by combining the data from Poverty Probability Index and Oxford Poverty & Human Development Initiative. Through applying linear regression model, decision tree, random forest model, gradian boosting model, and neural network to analysis existing data, the paper assesses respectively the extent to which the factors matter and the efficacy of each model. Final advancing employs cross validation and grid research. Through analysis and comparison, the paper concludes that generally, gradient boosting is the model with the highest accuracy for predicting poverty and education as the most influencing factor. The finale finishes upon the possible reason behind the factors.