Alzheimer's disease (AD) is a progressive neurodegenerative disorder that impairs memory, cognition, behavior, and other cognitive functions. Despite significant advances in understanding its molecular mechanisms, a definitive cure remains elusive. However, some treatments have the potential to slow disease progression if applied before brain damage occurs. Therefore, the identification of reliable biomarkers is critical for early diagnosis of AD and effective intervention. Recent advances in proteomics and the increased accuracy of machine learning algorithms have enhanced biomarker discovery and validation. In this study, we used a newly developed proteomic pipeline to analyse cerebrospinal fluid (CSF) to profile the proteome of AD patients, which included two different subgroups based on their CSF levels of tau. Then, machine learning was used to identify proteins that best classified the two subgroups of AD patients compared to non-AD controls. The resulting model, based on few CSF proteins, demonstrated high accuracy in predicting AD and differentiating patients with elevated or normal CSF tau levels. These protein classifiers, detectable in the preclinical stages of AD, were further validated in silico using larger, publicly available proteomic datasets, confirming their potential as early diagnostic tools.