COMPARISON OF NEURAL NETWORKS AND STATISTICAL-MODELS TO PREDICT GESTATIONAL-AGE AT BIRTH

The aim of this study was to produce models for tile prediction of hig h risk pregnancies, with particular emphasis on pre-term delivery. Neu ral network and logistic regression models have been developed utilisi ng pregnancy and delivery data spanning a period of seven years. Five input factors were used as explanatory variables: age, number of previ ous still births, gestational age at first clinical assessment, diabet es and a measure of socio-economic status. There was little difference between average model performance for the two techniques: optimal neu ral network performance was achieved with a fully connected feed forwa rd network comprising a single hidden layer of three nodes and single output node. This produced a Receiver Operating Characteristic (ROC) c urve area of 0.700. The ROC area for logistic regression models was 0. 695, The performance of these models reflected weak associations withi n the data. However, performance is encouraging given the relatively l imited number of predictive inputs.