HYBRID ENSEMBLE LEARNING APPROACHES FOR HIGH-ACCURACY DEMENTIA DETECTION: INTEGRATING DEEP LEARNING MODELS

Abstract

The main neurological disease called dementia impacts millions globally because its progressive nature combined with a complicated diagnostic procedure, creates significant diagnostic obstacles. The combination of neuroimaging with clinical evaluation leads to costly and variable time-consuming diagnostics that reduce the early detection ability of dementia. A diagnostic accuracy improvement model used combination techniques between CNNs and LSTM networks and SVMs to create an ensemble learning system. Recent research indicates deep learning holds promise but its application remains limited by overfitting problems and narrow scope. The proposed ensemble model performs better than previous approaches because it reaches 95.1% training and 92.8% validation accuracy rates. The ensemble model outperforms CNN and LSTM detection systems according to comparative analysis because it reaches 93.5% sensitivity and 91.2% specificity performance levels. The training model shows progressive reduction of loss values through downward movement until it completes its training cycle with final epoch loss levels at 0.2 training and 0.3 validation which ensures robust learning without overfitting. The ensemble approach successfully detected dementia cases which improves diagnosis methods while enhancing medical choice-making processes. New research will improve existing datasets through genetic and linguistic biomarkers while benefiting from Explainable AI (XAI) for greater interpretability in the analysis method. Research proves that medical institutions should use AI diagnostic tools in their practice because these systems deliver affordable scalable tests that diagnose dementia effectively.