Researching Machine Learning Methods for Diagnosing Women's Health

This article reviewed the existing methods for solving the problem (decision trees, neural network training), identified their advantages and disadvantages. The proposed method differs from the existing ones in that in order to identify the most significant signs affecting the assessment of the probability of getting a disease, the selection of signs was carried out on the basis of a genetic algorithm, as well as comparison and identification of effective methods of neural network learning. As a result of the comparison, the backpropagation Bayesian regularization method was chosen, since it showed the smallest recognition error with a relatively short training time. Thus, a neural network was trained to predict cervical cancer in women.