Implementação do método de regressão logística na classificação de exames por espectrometria de massa quanto à presença de câncer do ovário

Abstract

This work aims to implement the logistic regression method, a simple 01 single-layer Artificial Neural Network, to classify the results of mass spectrometry exams in 02 diagnosis classes: with or without ovarian cancer. An “OvarianInputs” database was used with data from 216 patients examined with ion intensities corresponding to 100 specific load-mass values and the “OvarianTargets” database with diagnostic results for network training purposes neural. The k-fold cross validation was used in 5 randomized folders to assess the average accuracy of the model. The confusion matrix obtained from the classification of the elements of the test set of each folder was used. The algorithm responsible for this implementation was developed using Python language libraries and compared with the results obtained from the mathematical formulation of the model in MATLAB software, reaching an average accuracy of 93.03% in both implementations.