Improving Early Alzheimer's Detection with Machine Learning and Feature Engineering on Blood Plasma Proteomics

Early detection of Alzheimer's disease is vital for timely care and treatment. This paper explores detecting AD stages from blood plasma proteomics data. The Alzheimer's Disease Neuroimaging Initiative dataset, with 146 plasma protein features, is analyzed using classifiers such as support vector machines, random forests, and gradient boosting. The models are optimized via grid search and 5 -fold cross-validation. Initial results show that gradient boosting achieves a test accuracy of 83.1 %. We apply data scaling and feature engineering methods such as mutual information, locality preserving projections, and principal component analysis to improve performance. The enhanced support vector machine model with mutual informationbased feature selection achieves an 84.5 % test accuracy using 98 key protein features. Six proteins are commonly selected by all models, suggesting their significance in AD detection. The study demonstrates the potential for accurate and non-invasive blood-based biomarkers in early AD diagnosis.