Comparative Sensitivity Performance of the Discriminant Function and Logistic Regression under Different Training and Test Samples for Predicting Birth Outcomes

Abstract

Population increases with time through birth, and researchers have often used either Logistic regression model or Discriminant analysis to study and classify birth outcomes. In this paper, the authors sought to investigate the sensitivity of the two methods used separately to explain and classify birth outcomes under different training and test samples. Out of 5000 birth outcomes data comprising of 1250 stillbirth cases and 3750 live births and with four test samples (50%, 40%, 30% and 25%). The Discriminant Analysis averagely correctly classified 89.8% of birth outcome cases compared to 82.4% for the logistic regression. The Discriminant analysis on the average correctly predicted 94.2% of live births compared to 83.1% for the Logistic regression. On stillbirth, 75.7% and 80.9% success rates were recorded for Discriminant Analysis and Logistic regression respectively. All predictors (Maternal Age, Gestational period, fetus weight, parity and Gravida) were statistically significant (p-value < 0.01) in determining birth outcomes of pregnancies in both methods. The results showed that, both techniques are almost similar in predicting birth outcome. However, the Discriminant analysis is preferred for the 25% and 50% test samples whiles, the logistic regression performed well under the 30% and 40% test sample data.