<![CDATA[Pregnancy-related complications require timely recognition, while symptom patterns may vary among patients. This study develops a decision-support model based on a backpropagation artificial neural network to classify six pregnancy-related conditions from symptom data. The dataset contains 172 records represented by 17 input variables and six output classes: normal pregnancy, early pregnancy, hyperemesis gravidarum, preeclampsia/eclampsia, hydatidiform mole, and ectopic pregnancy. Experiments evaluated training-test proportions of 90:10, 80:20, 70:30, 60:40, and 50:50, in addition to a resubstitution experiment using all records. The selected network used 17 input neurons, 50 hidden neurons, six output neurons, a sigmoid activation function, learning rate 0.1, target error 0.01, and a maximum of 50,000 epochs. Training accuracy reached 100% in the reported scenarios, whereas independent test accuracy ranged from 71.43% to 85.71%. The best holdout result was obtained with the 80:20 split, recognizing 30 of 35 test records (85.71%). These findings indicate that backpropagation can support multiclass pregnancy-condition classification, but further stratified validation and external clinical evaluation are required before practical diagnostic use.]]>
Copyrights © 2024
