Accurate surgery time prediction (ASTP) strategy based on artificial intelligence techniques

Source Domain: www.nature.com

The proposed Accurate Surgery Time Prediction (ASTP) framework consists of two layers: a preprocessing layer and a prediction layer.
The preprocessing layer uses techniques such as one-hot encoding, Mixed-Scaling, and methods to determine feature importance using LSTM and Random Forest, followed by feature ranking.
The prediction layer utilizes Histogram Gradient Boosting Regression (HGBR) to predict surgery duration based on the ranked features.
The LSTM model applies SHAP to quantify the importance of features, while Random Forest uses permutation importance for feature ranking to ensure robustness.
The HGBR model quantizes continuous features into bins, evaluates splits at bin boundaries, conducts gradient boosting, and averages predictions in log space for stable outputs.
The ASTP framework was evaluated on two datasets: the Nile Hospital dataset and the MOVER dataset, using metrics like mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R²).
HGBR achieved the best performance on both datasets, demonstrating lower mean absolute error and higher R² compared to other models like ANN, LSTM, GRU, and LSTM + GRU hybrids.
Statistical analysis confirmed that the HGBR-based ASTP model significantly outperformed other comparative models.

You may have missed