Deep ensemble learning for intelligent healthcare computing: A case study of Alzheimer’s disease

The growing popularity of deep learning (DL) in recent years has encouraged researchers to diversify their applications further. The limitations and shortcomings of an individual model are subdued through ensemble learning (EL), which combines the predictions of multiple models that are trained separately, thereby improving the overall accuracy and robustness. Deep ensemble learning (DEL) models leverage the combined diversity of different deep learning models. This paper provides an overview of traditional, novel, and state-of-the-art deep ensemble methods for application in Alzheimer's disease (AD) and other intelligent healthcare applications, including bagging, boosting, stacking, homogeneous/heterogeneous ensembles, explicit/implicit ensembles, negative correlation-based deep ensemble models and decision fusion. For this research study, an extensive exploration was conducted across prominent academic databases, including Google Scholar, ProQuest, DBLP, Science Direct, MDPI, IEEE Xplore, and Springer. The investigation encompassed a meticulous search for literature between 2018 and 2023 to ascertain the study's most current and relevant data. The results are presented through various methodologies, including flow charts, graphs, figures, and comparative tables, ensuring a comprehensive and visually accessible representation of the findings. This survey paper presents performance results from diverse ensemble methods applied to deep learning models. This reveals significant performance enhancements on specific datasets and model combinations, showcasing the impactful role of ensembling in surpassing individual model outcomes. Our findings also highlight nuanced correlations between ensemble techniques and data characteristics, offering actionable insights for implementing optimized ensemble-based deep learning models in clinical settings. This novel contribution underscores our paper's advancement in Alzheimer's detection methodologies, uniting comprehensive data analysis, ensemble effectiveness, and valuable considerations.