Disease prediction via Bayesian hyperparameter optimization and ensemble learning.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Early disease detection significantly improves patient survival rates.
• Identifying predictive features for conditions like breast cancer (BC) and cardiovascular disease (CVD) is crucial for timely intervention.
• Machine learning (ML) offers powerful tools for analyzing complex health data to identify early disease markers.

Purpose of the Study:

• To conduct a comparative analysis of various ML systems for predicting BC and CVD.
• To identify key early-stage predictive features for BC and CVD using feature importance ranking.
• To evaluate the stability of different hyperparameter optimization methods in ML models.

Main Methods:

• Comparative analysis of ML algorithms including Extreme Gradient Boosting (XGBoost).
• Utilized Bayesian hyperparameter optimization, grid search, and random search.
• Employed feature importance ranking to identify significant predictive variables.
• Validated models on BC and CVD datasets using metrics like accuracy and sensitivity.

Main Results:

• XGBoost achieved 94.74% accuracy and 93.69% sensitivity for BC prediction.
• Mean cell nucleus value from Fine Needle Puncture (FNA) images was the top BC predictor.
• XGBoost achieved 73.50% accuracy and 69.54% sensitivity for CVD prediction.
• Systolic blood pressure emerged as the most critical feature for CVD prediction.
• Bayesian optimization demonstrated superior stability compared to grid and random search.

Conclusions:

• ML models, particularly XGBoost, show high efficacy in predicting BC and CVD.
• Specific features like mean cell nucleus value (BC) and systolic blood pressure (CVD) are vital early indicators.
• Bayesian hyperparameter optimization enhances ML model stability and reliability.
• This research highlights the potential of ML for early disease screening and risk stratification.