Electroencephalography (EEG) is a non-invasive neurophysiological measurement method that allows monitoring the electrical activity of the cerebral cortex and is widely used in the diagnosis of neurological diseases. In this study, EEG-based biomarkers were used to discriminate between Alzheimer's disease, frontotemporal dementia, and cognitively healthy individuals. A total of 22 features were extracted in the signal processing stage, and then this number was reduced to 12 by applying a feature selection method based on the ReliefF algorithm to improve the classification performance. The selected features were evaluated in both binary and multiclass classification scenarios to reveal the discriminative differences between Alzheimer's disease, frontotemporal dementia and healthy control group. According to the findings, in the multiclass classification task, the Fine Decision Tree algorithm achieved the highest accuracy rate of 99.7% when all features were used. In distinguishing cognitively normal individuals from individuals with Alzheimer's disease, both the Fine Decision Tree and Cubic Support Vector Machine algorithms achieved 100% accuracy with all and selected feature sets. To prevent overfitting and evaluate generalization performance, k-fold cross-validation was applied. Feature selection and model parameter tuning were performed only on the training folds; the test folds were not included in these processes. This approach prevents information leakage and provides reliable performance estimation. This finding demonstrates that EEG-based biomarkers, when combined with appropriate machine learning methods, can be transformed into effective tools that provide high reliability and accuracy in clinical decision support systems.