Artificial Neural Network techniques to distinguish children with Fetal Alcohol Spectrum Disorder from psychometric data

In recent decades, one of the most used technique is artificial neural networks (ANN), since their learning is based on a set of connections, it is transparent to the user and the result has managed to solve complex problems in the medical field. The study implemented algorithms of ANN using input data from a battery of psychometric tests. This battery assesses multiple domains of attention and executive functioning, memory and learning, sensorimotor functioning, social perception, language, and visual-spatial processing. We attempt to explore how accuracy is the use of ANN for the prediction of children with Fetal Alcohol Spectrum Disorder (FASD). We implemented the ANN with a configuration of three layers, 20 neurons in the input layer, 25 neurons in the hidden layer, and two neurons in the output layer. We studied the accuracy of the model in training and testing, also the confusion matrix of the model. Using our machine learning approach, we have trained the ANN model to predict children/adolescents with FASD with accuracy ranging from 75.5% in testing data. These results suggest that the ANN approach is a competitive and efficient methodology to detect and differentiate the cognitive neurodevelopmental consequences of prenatal alcohol exposure. However, we could not recommend the use of this technique for diagnosis FASD if the model does not improve accuracy.