Quokka swarm defensive optimization-based ensemble deep learning for disease prediction using remote monitoring data

Abstract

Accurate disease prediction from patient monitoring data is essential for enabling timely clinical intervention. However, existing approaches often exhibit limited generalization and reduced robustness to noisy or incomplete physiological signals. To address the challenges, this article proposes a novel approach called Quokka swarm defensive optimization algorithm-based Ensemble deep learning (QSDO_Ensemble DL) for disease prediction. At first, physiological signals, like II, III, AVL, AVF, and Arterial Blood Pressure (ABP), are used for anomaly detection, and it is carried out by using a weighted average method. In parallel, the same set of health parameters is used to predict Pulmonary Arterial Pressure (PAP), and it is done through an Adaptive Hybrid Attention Network (AHANet). Subsequently, the outputs from anomaly detection and PAP prediction stages are fed into a disease detection system. Here, disease prediction is done using Ensemble deep learning (DL) classifiers, including ResNeXt, AHANet, and Shuffle Attention Network (SA-Net). In addition, the hyperparameters of the ensemble DL are tuned by QSDO, which combines the Quokka swarm optimization (QSO) and Offensive Defensive Optimization (ODO). Experiments are conducted on the MIMIC-II database, which is used for model training and evaluation. Moreover, the QSDO_Ensemble DL achieved an Accuracy of 96.90%, a True Positive Rate (TPR) of 95.87%, and a False Positive Rate (FPR) of 7.22%. The results indicate that QSDO_Ensemble DL provides robust and reliable disease prediction from physiological monitoring signals.