A Smart IoT-Based System for Early Prediction of Parkinson’s Disease Using Swarm-Assisted Convolutional Neural Networks

The ever-increasing volume of patient health data presents persistent challenges for automated healthcare systems in accurately classifying and diagnosing various ailments. The complexity of training on diverse variables and redundant data often reduces classification accuracy. Hybrid models based on convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and support vector machines (SVMs) have been investigated for disease classification, such as Parkinson’s disease. However, the selection of ideal cores and extreme parameters has a significant impact on their effectiveness. To address these shortcomings, this research presents an enhanced IoT-based model for the analysis of medical data. An evolutionary deep belief network (EDBN) is used to collect and process the patient health data, firstly optimising the learning variables using the adaptive predator–prey optimisation algorithm, in contrast to the traditional deep belief network (DBN). The EDBN can extract pertinent features with lower computational requirements. The hybrid classifier, which combines CNN, LSTM, and swarm-based deep feature self-organiser (SDFSO), is then used to classify diseases with high accuracy. Normalisation of Dropouts and hyperparameter optimisation are used to prevent overfitting, as well as enhance convergence. The extensive review of publicly available medical datasets reveals that the proposed EDBN with CNN–LSTM–SDFSO model is more effective, with an accuracy of 99.14%, which is significantly more effective than traditional methods. This IoT–fog–cloud model provides a viable value-added system for the real-time and early detection of Parkinson’s disease under strained healthcare conditions. Further analyses established high sensitivity (97.8%), specificity (99.08%), precision (97.8%), recall (99.08%), F1-score (99.7%), and AUC, which contributed to its clinical applicability.