An Explainable Clinical Decision Support Framework Using Adaptive Marine Predators Algorithm-Optimized Improved CNN For Accurate Weighted Fuzzy Production Rule Extraction

Abstract

Clinical Decision Support Systems (CDSSs) need a predictive accuracy as well as interpretability to assist clinicians in disease diagnosis and risk assessment. Traditional Convolutional Neural Networks (CNNs) manage to capture complex nonlinear relations but suffer from opacity in terms of outcomes, are sensitive to initialization of hyperparameters and cannot be utilised for explainable decision-making in healthcare. This research presents a novel explainable as well as hybrid framework to integrate an Adaptive Marine Predators Algorithm (AMPA)-optimized Improved CNN with the extraction of Weighted Fuzzy Production Rule (WFPR) to bridge predictive accuracy and interpretability. The upgraded CNN employs convolutional filters operating at different scales, as well as residual connections that allow for the passage of gradients, which address the problem of vanishing gradients. In addition, it uses attention to weight features that are clinically relevant. The Adaptive Marine Predators Algorithm is put into play in the weight and hyperparameter optimization of the CNN. The model uses adaptive Brownian and Levy motion to ensure global convergence and to avoid local optima or loss in high-dimensional search spaces. The post Adaptive Marine Predators Algorithm EM Asymmetrical Black Hole Optimization appeared first on Javatpoint. The outputs of CNNs are subjected to fuzzy membership modelling. Therefore, interpretable WFPRs are inferred, transforming underlying complex learned representations into easily understandable decision rules by clinicians. This framework was tested on various clinical datasets which showed increased classification accuracy, strong training stability as well as better rule interpretability. Overall, the CDSS architecture is transparent and scalable and can ensure reliable health care insight.