Imbalanced-class binary classification for Santander Customer Satisfaction Prediction based on 200 variables with Keras. AUC score of 0.8 https://www.kaggle.com/c/santander-customer-transaction-prediction/
Simple neural network architecture on Keras of 2 dense layers with size 50, with PReLu activation function (the last layer is sigmoid), dropouts and layer normalizations. Imbalanced classes require different weights for 2 outcomes.