Alzheimer's disease (AD) is a neurodegenerative disease affecting an estimated 6.5 million Americans and is expected to double in the next thirty years. The only treatments that patients can receive for the disease are to slow the disease progression, which is more effective if the disease is diagnosed early. A popular form of early diagnosis used by doctors is to analyze the brain's electroencephalogram (EEG). EEG measures electrical activity by placing electrodes on the scalp or surgically placing electrodes inside the brain. Given the electrical activity measurements of the brain, high and low-frequency signals are traditionally filtered during preprocessing as previously believed to be noise or irrelevant information. However, we believe this filtered data contains valuable information that can help doctors diagnose Alzheimer's patients earlier. We use novel machine learning models, such as transformers, and additionally use metrics such as accuracy, precision, recall, and F1 score to evaluate the model performance. Our methods result in higher accuracy compared to current machine learning methods. Our best-performing transformer resulted in an accuracy of 0.846, while the best-performing non-transformer model resulted in an accuracy of 0.564.