Alzheimer’s disease (AD) is the most prevalent type of dementia in people over the age of 60, initially affecting the hippocampus, and causing difficulty in remembering recent events. It then spreads throughout the brain, causing progressive cognitive impairment. This research describes a method for the selection of specific Electroencephalography (EEG) channels in AD patients and healthy Control Subjects (CN) to identify the brain areas that provide more information about these classes. Time, frequency, time-frequency, and fractal dimension features were extracted for classification, followed by selection using One-Way Analysis of Variance (ANOVA). We compared five supervised learning classifiers: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and Multilayer Perceptron (MLP) using Leave-One-Person-Out (LOSO) Cross-Validation (CV). Classifiers were evaluated with Receiver Operating Characteristic (ROC) curve and the Area Under Curve (AUC). Afterwards, a Mann-Whitney-Wilcoxon (MWW) test of the best classification models was performed. We also proposed an approach for subject exclusion based on topographic analysis of the electrical activity of the signals. Our results indicate that the Random Forest algorithm outperforms the others, achieving an Accuracy of 91.07(± 2.1)% and an F1 score of 91.22(± 2.8)%, underscoring its potential for improving the diagnosis of patients with AD.