Models | Hyperparameters | Algorithm characteristic |
---|---|---|
GBDT | n_estimator = 100, max_depth = 3 | GBDT is the state of the art of machine learning model with extremely high accuracy |
RF | n_estimator = 100 | 1. Good regression performance for high dimensional data. 2. No easy overfitting. 3. Fast Training speed, easy to make parallel method |
BPNN | learning_rate = ,hidden_layer = 2, hidden_size = [128,128], drop_out = 0.1 | 1. Strong robustness and fault tolerance. 2. Can fully approximate any complex nonlinear relation |
TRNN, LSTM, GRU | learning_rate = 0.0002–0.0005, hidden_layer = 2, hidden_size = [128,128], drop_out = 0.1 | TRNN, LSTM, GRU are the most popular deep learning frameworks for processing time series data |
A-TRNN-PC, A-LSTM-PC, A-GRU-PC | learning_rate = 0.0001–0.0003, heads = 2,hidden_layer = 2, hidden_size = [128,128], drop_out = 0.1 | \ |