Predictability of Some Pregnancy Outcomes Based on SVM and Dichotomous Regression Techniques

The objective of this study is developing a forecasting system for some childbirth outcomes, based on an input pattern of instrumental and anamnestic parameters detected at 37th week of pregnancy. The study stems from the need to be able to predict what to expect during labor and childbirth, while discovering new knowledge from the evidence of the data (data mining). Outcomes to predict concern: underweight newborn, post partum bleeding, need for artificially induced birth, necessity of cesarean birth. The predictors parameters are a total of 58 dichotomous inputs grouped into 4 categories: pre-conception risk factors, obstetric risk factors, risk factors associated with pregnancy, ultrasound parameters. The training database is populated with 420 patients, each with a single follow-up. Best leave one out cross-validation results were achieved in the estimation of underweight (ROC point chosen, sensitivity 0.69 - specificity 0.88).