[Study on a back propogation neural network-based predictive model for prevalence of birth defect]

Area of Science:

• Epidemiology
• Artificial Intelligence
• Biostatistics

Background:

• Birth defects pose a significant public health challenge.
• Accurate prediction of birth defect prevalence is crucial for effective prevention strategies.
• Traditional statistical methods may have limitations in handling complex epidemiological data.

Purpose of the Study:

• To assess the predictive accuracy of a back propagation (BP) artificial neural network for birth defect prevalence.
• To explore the utility of BP networks in providing insights for birth defect prevention.
• To compare the performance of BP networks against conventional statistical methods in epidemiological prediction.

Main Methods:

• Utilized birth defect data from Shenyang (1995-2005) as a training dataset.
• Employed the MATLAB 6.5 Neural Network Toolbox to train and simulate the BP Artificial Neural Network.
• Performed predictions for future prevalence rates using historical data.

Main Results:

• BP network achieved a fitting average error of 1.34% and a prediction average error of 1.78% when forecasting 2004-2005 rates using 1995-2003 data (RNL=0.9874).
• With 1995-2005 data, the network predicted 2006-2007 prevalence rates of 11.00% and 11.29% with a fitting average error of 0.33% (RNL=0.9954).
• Demonstrated high precision in predicting birth defect prevalence rates.

Conclusions:

• BP artificial neural networks exhibit superior prediction precision compared to conventional statistical methods.
• BP networks are versatile, with no limitations on data type or distribution, making them suitable for epidemiological prediction.
• This technology offers a powerful and adaptable approach for predicting and potentially preventing birth defects.