From: Improving mortality forecasting using a hybrid of Lee–Carter and stacking ensemble model
Model | Parameter | Value |
---|---|---|
Neural networks (NN) | Hidden layers configuration | (60, 50) |
Cross-validation | nfold = 5 | |
Fold assignment method | Fold assignment = modulo | |
Early stopping | Stopping round = 50 | |
Early stopping metric | Stopping metric = RMSE | |
Tolerance | Tolerance = 0 | |
Number of epochs | Epochs = 50 | |
Activation function | Activation = rectifier | |
Extreme gradient boosting (XGBoost) | Number of trees | ntress = 5000 |
Cross-validation | nfold = 5 | |
Fold assignment method | Fold assignment = modulo | |
Early stopping | Stopping round = 50 | |
Early stopping metric | Stopping metric = RMSE | |
Tolerance | Tolerance = 0 | |
Learning rate | learn_rate = 0.1 | |
Subsample rate | Sample rate = 0.8 | |
Maximum tree depth | max_depth = 5 | |
Random forest (RF) | Number of trees | ntrees = 1000 |
Number of splits (mtries) | mtries = 2 | |
Cross-validation | nfold = 5 | |
Fold assignment method | Fold assignment = modulo | |
Early stopping | Stopping round = 50 | |
Early stopping metric | Stopping metric = RMSE | |
Tolerance | Tolerance = 0 | |
Subsample rate | Sample rate = 0.8 | |
Maximum tree depth | max_depth = 30 | |
Generalized linear model (GLM) | Alpha | Alpha = 0.1 |
Cross-validation | nfolds = 5 | |
Early stopping | Stopping round = 50 | |
Early stopping metric | Stopping metric = rmse | |
Tolerance | Tolerance = 0 |