Alzheimer’s Detection via Cepstral Domain Feature Fusion and Mann-Whitney U-Test Statistical Filtering

In this paper, we propose a robust approach for early detection of Alzheimer’s Disease (AD) using advanced cepstral speech features, fused to capture diverse acoustic signatures. The Mann-Whitney U-test is employed to filter out non-informative features, improving real-world applicability and reducing overfitting. We systematically compare Random Forest, Decision Tree, and Support Vector Machine classifiers, achieving an accuracy of 85.6% using only the most significant features, drawn from MFCCs, GTCCs, LPCCs, and others. The method addresses data sparsity challenges in clinical research by favoring domain-relevant indicators while maintaining computational feasibility. The results confirm that subtle vocal cues, captured using affordable devices, are promising biomarkers for AD screening, especially in resourcelimited or telehealth environments. By highlighting both practical implementation and scalability, our study contributes to the development of efficient, non-invasive diagnostic protocols. This, in turn, supports earlier intervention and improved patient outcomes. The framework thus paves the way for future multimodal expansions.