IoT-Enabled Intelligent Diagnostic Framework Using Deep Learning and Multi-Feature Fusion Optimization

Accurate and early identification of diseases remains a major challenge in modern healthcare, where diagnostic delays can lead to critical outcomes. The shortage of medical specialists and the growing demand for rapid and reliable diagnosis highlight the need for intelligent and automated decision-support frameworks. This paper introduces a novel Medical Intelligence for Accurate Disease Detection using Deep Learning and Feature Fusion Strategies (MIADD-DLFFS) model designed for IoT-enabled smart healthcare systems. The proposed system integrates advanced deep learning models with cloud-assisted communication to enable real-time data analytics and automated diagnostic decision-making. A bilateral filtering mechanism enhances the input data quality, while a hybrid fusion of three optimized neural architectures - SqueezeNet, enhanced InceptionResNetV2, and improved EfficientNetB7 - ensures robust and multiscale feature representation. For classification, an Enhanced Deep Q-Network (EDQN) is employed to deliver accurate diagnostic predictions, further optimized using the Barber Optimization Algorithm (BaOA) for adaptive hyperparameter tuning. The combination of deep learning and IoT infrastructure enables continuous health monitoring, remote diagnosis, and secure alert dissemination through cloud-based services.The proposed framework has three contributions to methodologies (i) a mid-level multiscale feature fusion pipeline which is a combination of SqueezeNet, an Enhanced InceptionResNet-V2 and an Improved EfficientNet-B7 (ii) a reinforcement-based Enhanced Deep Q-Network (EDQN) designed to work with dynamic classification; and (iii) an adaptive Barber Optimization Algorithm (BaOA) that works with hyper-parameter self-tuning. The study proposes an intelligent diagnostic framework that utilizes multiscale deep feature fusion with adaptive optimization to identify a disease precisely with the help of the IoT. The framework integrates lightweight convolutional models with deep convolutional models together with reinforcement-based classification, so that it can provide robust, scalable, and real-time healthcare diagnostics with better generalization performance. Experimental evaluation based on five-fold cross-validation of the ISIC dataset showed an average accuracy of 94.8 per cent that demonstrates that the generalization process is better and, consequently, supports the effectiveness and feasibility of the MIASCD-DLFFS architecture in the device of smart healthcare diagnostics in real-time.