<![CDATA[This study comparatively evaluates the performance of Bridge, LASSO, and Ridge regression methods in predicting body fat percentage based on anthropometric measurements. Exploratory analysis indicated strong relationships between body fat percentage and abdominal and hip circumferences, with evidence of multicollinearity among the predictors. The ordinary least squares (OLS) model identified significant predictors but violated several classical assumptions, particularly multicollinearity, heteroskedasticity, and autocorrelation. Therefore, penalized regression methods were applied to obtain more stable parameter estimates and improve predictive performance. Model performance was assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), coefficient of determination (R²), adjusted R², Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Variance Inflation Factor (VIF). The results show that Bridge regression with γ = 0.5 achieved the best predictive performance, producing the lowest MSE and AIC values. LASSO regression provided the simplest model by eliminating irrelevant predictors and achieved the lowest BIC and VIF values, while Ridge regression produced stable estimates but did not outperform the other methods. Overall, Bridge regression is recommended for prediction, whereas LASSO is preferable when model simplicity and interpretability are prioritized.]]>
Copyrights © 2026
