World Bank Poverty Prediction
Driven Data Competition
https://www.drivendata.org/competitions/50/worldbank-poverty-prediction/
Summary: Artificial neural networks can be used to predict which features in a dataset predict a household's poverty.
Data: 2018 World Bank household-level survey training data with 8200 observations, 343 features (reduced to 4 with PCA) and 1 target variable (poverty).
Results: k-Nearest Neighbor precision 0.52, recall 0.52, F1 score 0.52; stochastic gradient descent precision 0.52, recall 0.52, F1 score 0.52, mean log loss 15.68; multilayer perceptron 114 neurons, 2 layers, and lbfgs solver precision 0.85, recall 0.85, F1 score 0.85, mean log loss 12.22,; multilayer perceptron 342 neurons, 4 layers, and lgfgs solver precision 1.0, recall 1.0, f1 score 1.0, mean log loss 9.99 e-16).
IDB Costa Rican Household Poverty Level Prediction
Kaggle Competition
https://www.kaggle.com/c/costa-rican-household-poverty-prediction
Summary: Machine learning classification techniques can be used to predict which features in a dataset predict a household's poverty.
Data: 2018 International Development Bank household-level survey training data with 9557 observations, 143 features and 1 target variable (poverty level).
Results: Random Forest had macro F1 score = 0.9976