Enhanced prediction of spine surgery outcomes using advanced machine learning techniques and oversampling methods

Area of Science:

• Computational biology
• Medical informatics
• Machine learning in healthcare

Background:

• Accurate prediction of spine surgery outcomes is crucial for effective treatment planning.
• Machine learning offers potential for improving the predictive accuracy of surgical success.
• Existing models may require enhancement to handle data complexities and imbalances.

Purpose of the Study:

• To develop and evaluate an enhanced machine learning approach for predicting spine surgery success.
• To investigate the impact of advanced oversampling techniques and grid search optimization on model performance.
• To identify the most effective machine learning model for classifying spine surgery outcomes.

Main Methods:

• Applied various machine learning models (GaussianNB, ComplementNB, KNN, Decision Tree) to a dataset of 244 spine surgery patients.
• Utilized oversampling techniques including RandomOverSampler and SMOTE to address potential data imbalances.
• Employed grid search optimization for KNN and Decision Tree models to enhance predictive capabilities.

Main Results:

• The K-Nearest Neighbors (KNN) model, enhanced with RandomOverSampler and SMOTE, achieved the highest performance.
• Achieved accuracy up to 76% and an F1-score of 67% with the optimized KNN model.
• Grid-searched optimized KNN and Decision Tree models demonstrated significant improvements in accuracy and F1-score.

Conclusions:

• Advanced machine learning and oversampling methods can significantly improve spine surgery outcome prediction.
• Model optimization and careful variable selection are key to achieving high predictive performance.
• This approach can serve as a valuable tool for healthcare professionals in surgical decision-making.