Importance of Data Preprocessing in Artificial Neural Network Training for Medical Diagnosis

Artificial intelligence and machine learning techniques have been widely used to solve problems in multiple areas. The use of classification algorithms allows to automation of some daily life tasks, such as medical diagnosis. Specifically, the use of artificial neural networks oriented to classification tasks has shown that they produce good results for diagnosing many diseases. In this chapter, we analyze the impact of preprocessing techniques on backpropagation artificial neural networks applied to medical diagnosis. The preprocessing techniques studied consider normalization, standardization, and dataset sorting. On the other hand, we show a principal component analysis of the attributes in the dataset and explore the details of the data to identify characteristics that could help to select one specific type of preprocessing. The experimentation was carried out with five datasets of different diseases obtained from electronic machine learning databases. With the results obtained, it is possible to identify two significant points: first, the importance of data analysis and the impact on classification certainty by applying different preprocessing criteria, and second the effect on learning by studying some key parameters that determine training certainty.