The search for non-invasive biomarkers to support the early detection of dementia is a key priority in neurodegenerative disease research. This study explores the potential of steady-state visual evoked potentials (SSVEPs) recorded via electroencephalography (EEG) in combination with machine learning to distinguish between cognitively healthy individuals and patients with dementia. EEG signals were recorded from nine participants during visual stimulation at four different frequencies (8, 15, 27, and 60 Hz). A total of 38 features were extracted, including signal-to-noise ratio (SNR) metrics and spectral amplitudes at harmonic frequencies. Feature selection was performed using manual and automatic methods, resulting in eight variable subsets. Classification performance was evaluated using multiple machine learning algorithms, including k-Nearest Neighbors (kNN), Support Vector Machines (SVM), Logistic Regression, Decision Trees, and Neural Networks. Subject-wise leave-one-out cross-validation (LOSO) was used to avoid overfitting. The highest accuracy was obtained with kNN (up to 80%) using a reduced feature set. Frequencies of 8 Hz and 15 Hz yielded the most discriminative features, while 60 Hz was less effective in eliciting SSVEP responses. Results suggest that SSVEP-based features, particularly spectral amplitude and SNR, hold promise as complementary biomarkers for cognitive impairment screening. Further studies with larger cohorts are needed to validate these findings and support clinical applications.