POVERTY PREDICTION USING MACHINE LEARNING APPROACH
Pa Pa Min, Yen Wen Gan, Siti Nursyuhada Binti Hamzah, Thian Song Ong, Md Shohel Sayeed
Abstract
The incidence of poverty is not a taboo topic. In fact, it happens in every country worldwide where the policymakers and governments struggle to reduce their country’s poverty rate. However, the existing ways of finding the right targeted impoverished group to provide economic aid are often flawed because of multiple issues such as data transparency and incorrect, redundant, or imbalanced data. Even though there are solutions available for the problems, some of them are expensive and require a lot of human intervention efforts. Thus, there is an alternative solution to use machine learning approach to forecasting and providing cost efficiency and effective solution to address poverty issues. This research utilizes machine learning model which comprises several pre-processing techniques for data cleaning, feature engineering and then followed by three different regression methods for predictive analytics. The proposed machine learning model is evaluated using Costa Rican household poverty dataset and the experimental results indicate that out of three different regression models used, Random Forest Regressor has the best result in terms of R2 and RMSE score which are 0.9462 well-fitted and 0.2591 error rate, respectively. Moreover, we analyzed the importance of individual features for poverty prediction using SHAP value based on summary plot analysis.