Full data and model exploration notebook to predict medical ventilator pressure based on (patient-specific) past ventilator state data. Created for the Google Brain - Ventilator Pressure Prediction competition hosted by Kaggle.
The original notebook (with version history and connection to the relevant data sources) is availabile on Kaggle Notebooks.
Contents:
- initial exploration of time-series data,
- regression models using time-series data:
- linear regression,
- SGDR,
- vanilla dense neural network,
- LSTM, RNN, GRU,
- ensemble methods: boosting,
- hyperparameter tuning:
- keras_tuner.