This time we focus on predict the patient's mean-MAP (mean arterial pressure) and mean-HR (mean heart rate) given a key (for a patientid and a period of time).
The training data contains patient id, patient key, age, and other features (x1, x2, etc.) where we can't know the meaning through the column names.
We use python to do data analysis.
In the feature engineering part, we tried upsampling and downsampling, clustering, mean encoding/target encoding , aggregation and missing value imputation. In the modeling part, we tried using XGBoost, Random Forest, Light-BGM model.
The best validation score is 0.92711.