zscore Customized process - > tsvd & pca -> row- wise zscore Target High correlated features -> row- wise zscore predictions -> tsvd - > row-wise zscore BiGRU Layer Dense Layer Dense Layer Dense Layer Dense Layer BiGRU Layer Dense Layer NN 1 NN 2 Loss: Cosine similarity Hidden neurons:1800 GussianDropout:0.2 Activation:elu Initializer:Identity LR:0.001 Optimizer: Adam Concatenate:3 hidden layers pseudo labeling Loss: MSE Hidden neurons:1500 GussianDropout:0.1 Activation:swish LR: 0.0005 Optimizer: AdamW Target Zscore pseudo labeling LightGBM