Deep Ensemble of Classifiers for Alzheimer's Disease Detection with Optimal Feature Set

Machine learning (ML) and deep learning (DL) techniques can considerably enhance the process of making a precise diagnosis of Alzheimer's disease (AD). Recently, DL techniques have had considerable success in processing medical data. They still have drawbacks, like large data requirements and a protracted training phase. With this concern, we have developed a novel strategy with the four stages. In the initial stage, the input data is subjected to data imbalance processing, which is crucial for enhancing the accuracy of disease detection. Subsequently, entropy-based, correlation-based, and improved mutual information-based features will be extracted from these pre-processed data. However, the curse of dimensionality will be a serious issue in this work, and hence we have sorted it out via optimization strategy. Particularly, the tunicate updated golden eagle optimization (TUGEO) algorithm is proposed to pick out the optimal features from the extracted features. Finally, the ensemble classifier, which integrates models like CNN, DBN, and improved RNN is modeled to diagnose the diseases by training the selected optimal features from the previous stage. The suggested model achieves the maximum F-measure as 97.67, which is better than the extant methods like TSO = 72.39%, BMO = 78, SSA = 84.15%, GEO = 70.39%, and FFLY = 73.13%, respectively. The suggested TUGEO-based AD detection is then compared to the traditional models like various performance matrices including accuracy, sensitivity, specificity, and precision.