Comparison of classifiers when analyzing electroencephalographic signal behavior in comatose patients

In the present study, 75 EEG records of comatose patients from the Uberlândia Clinical Hospital of UFU were analyzed. These records were subdivided into three groups: Active group with 21 records, Clinical Death (OC) group with 42 records and Brain Death (OME) group with 12 records. At all EEG signs the Power Contribution Percentage (PCP) and Brain Power Variation (VPC) were the quantifiers calculated, and from them, the performance of the Logistic Regression and Support Vector Machine (SVM) classifiers were compared using the ROC curve tool for detecting which behaves best in a possible coma prognosis. It was observed that Logistic Regression better segmented comatose data, thus separating the Active outcome from the Death outcome.