Evolutionary methods for variable selection in the epidemiological modeling of cardiovascular diseases

Area of Science:

• Epidemiology
• Bioinformatics
• Machine Learning

Background:

• High-dimensional data presents challenges for epidemiologists.
• Effective variable selection is crucial for accurate predictive modeling.
• The Kuopio Ischemic Heart Disease Risk Factor Study (KIHD) dataset was utilized.

Purpose of the Study:

• To implement and evaluate an advanced evolutionary variable selection method.
• To compare this method against conventional stepwise selection.
• To identify key predictors for cardiovascular disease.

Main Methods:

• An evolutionary search-based variable selection method was developed.
• The method was applied to a dataset with 433 predictor variables.
• Performance was assessed using Generalized Linear Logistic Regression and Support Vector Machine models.

Main Results:

• The number of variables was reduced from 433 to 38.
• Predictive ability was preserved, with F-scores improving slightly post-selection.
• Results were validated using a 5-fold cross-validation procedure.

Conclusions:

• The evolutionary method effectively identifies relevant variables for cardiovascular disease prediction.
• A curated list of essential variables is provided for future epidemiological research.
• Variable multicollinearity allows for alternative predictor combinations without compromising model performance.