Çarpık hata dağılımları altında çoklu doğrusal regresyon modeli: fetal ağırlık tahmini üzerine bir vaka çalışması

Fetal weight estimation is an essential component of pregnancy care. Current weight estimation methods use parameters obtained from multiple linear regression with least squares or maximum likelihood under normality assumption. However, non-normal errors are very common in real life setting irrespective of the sample size. This work demonstrates the properties of least squares, ridge, maximum likelihood, and modified maximum likelihood estimators of the parameters of multiple linear regression model under skewed error distributions, namely skewed normal and generalized logistic. Performance metrics are obtained from a real-life sample of fetuses with ultrasonographic limb, abdomen and head measurements. Estimation and prediction performance are compared between available and new algorithms using standard error of the estimates, Akaike and Bayesian Information criteria and prediction errors.