Towards precise chronic disease management: A combined approach with binary metaheuristics and ensemble deep learning

Abstract

Chronic disease (CD) recognition involves identifying the existence or risk of CDs in individuals. CDs have chronic health illnesses categorized by slow progression and frequent reduction from intricate reasons. CDs comprise chronic respiratory diseases, heart disease, diabetes mellitus, and certain cancers. Earlier diagnosis is vital in handling CDs proficiently. Then, it permits lifestyle modifications, timely intervention, and medical services to avoid the progression of the disease and reduce its effect on their health. Recently, technical development, particularly in healthcare statistics and artificial intelligence (AI), has assisted in advancing sophisticated approaches and systems for CD recognition. These methodologies usually employ deep learning (DL) and machine learning (ML) models for investigating enormous databases, identifying patterns, and making predictions that rely on distinct health-related parameters. This study presents an accurate chronic disease detection and classification model using binary meta-heuristics with an ensemble deep learning (ACDDC-BMEDL) approach. The ACDDC-BMEDL methodology focuses on the procedure of average ensemble classifier with meta-heuristic-based feature selection (FS) and hyperparameter tuning processes. The ACDDC-BMEDL methodology uses a binary arithmetic optimization algorithm (BAOA) to choose better feature subsets. Additionally, the ACDDC-BMEDL methodology uses an average ensemble technique encompassing recurrent neural network (RNN), gated recurrent unit (GRU), and extreme learning machine (ELM) for classification procedure. The marine predator's algorithm (MPA) is employed for the hyperparameter tuning process. The experimental value of the ACDDC-BMEDL methodology was examined on 2 CD datasets. The performance validation of the ACDDC-BMEDL methodology portrays a superior value of 98.70% and 94.51% with recent methods concerning several metrics under Diabetes and HD datasets.