Sample-Pair Envelope Diamond Autoencoder Ensemble Algorithm for Chronic Disease Recognition

Chronic diseases are severe and life-threatening, their accurate early diagnosis is difficult. Machine-learning-based processes of data collected from the human body using wearable sensors a valid method currently usable for diagnosis. However, it difficult sensor systems to obtain high-quality large amounts meet demands diagnostic accuracy. Furthermore, existing feature-learning methods do not deal with this problem well. To address above issues, sample-pair envelope diamond autoencoder ensemble algorithm (SP_DFsaeLA) proposed. The proposed has four main components. Firstly, manifold neighborhood concatenation mechanism (SP_EMNCM) designed find pairs samples that close each other in neighborhood. Secondly, feature-embedding stacked sparse (FESSAE) extend features. Thirdly, staged feature reduction reduce redundancy extended Fourthly, sample-pair-based model single-sample-based combined by weighted fusion. was experimentally validated on nine datasets compared latest algorithm. experimental results show significantly better than representative algorithms achieves highest improvement 22.77%, 21.03%, 24.5%, 27.89%, 10.65% five criteria over state-of-the-art methods.