A Novel Personalized Random Forest Algorithm for Clinical Outcome Prediction.

Area of Science:

• Machine learning
• Biomedical informatics
• Personalized medicine

Background:

• Population machine learning models predict outcomes for patient groups but may fail for individuals.
• Personalized machine learning aims to improve prediction accuracy for each patient.
• Ensembles of decision trees are effective, but personalized ensemble models are under-explored.

Purpose of the Study:

• To introduce Lazy Random Forest (LazyRF), a novel personalized ensemble method.
• To evaluate LazyRF's performance against single and bagged decision paths and population-based random forests.

Main Methods:

• Developed Lazy Random Forest (LazyRF), an ensemble of bagged randomized decision paths.
• Optimized decision paths for individual patient predictions.
• Tested LazyRF on clinical and genomic datasets.

Main Results:

• LazyRF outperformed single and bagged decision paths in predictive performance.
• LazyRF demonstrated comparable predictive discrimination to population random forests.
• LazyRF generated simpler models compared to population random forests.

Conclusions:

• LazyRF is a promising personalized machine learning approach for predicting patient outcomes.
• Personalized ensembles can achieve high predictive performance while offering model interpretability.
• LazyRF advances the field of personalized medicine through tailored predictive modeling.