| Abstract |
In smart cities, Takayasu’s arthritis (TA) is the most dangerous rare cardiac vascular disease and is characterized by a variety of features. So, predicting the disease’s nature and effects with early treatment is difficult. During medical data analysis, the impact of disease rate is varied due to patient conditions and feature margins. Most existing models in smart cities fail to analyze the feature dimensions related to the disease. This leads to inaccurate disease identification because property features have no mutual dependencies to predict. So, a federated learning model is used to enhance detection accuracy and analyze medical data in a smart city. The increase in the non-feature relation does not create a mutual feature relation to take the impact margins of disease identification support weights into account. To resolve this problem, we propose a quantitative grey wolf optimized transfer learning (QGWOTL) model dependent on the multi-perceptron neural network (MPNN) to determine the nature of the disease so as to be able to provide earlier treatment. The preprocessing uses Z-score vector normalization (Z-SVN) to reduce the noise ratio. With this attention, the ideal medical margin disease rate is estimated by the cardiac-deficient impact rate (CDIR). The TA-disease-prone factor (TADF) creates the disease feature pattern (DFP) to marginalize feature weights. Based on the actual margins, both TADF and DFP ideal values are compared to identify the feature variation needed to create clusters of different features. Grey wolf optimized with support vector machine (GWO-SVM) is used to select features depending on marginalized cluster groups and class and risk classification depending on the LSTM-gated optimized with MPNN (LSTMG-MPNN). The results are categorized by risk based on the patient’s condition to prioritize earlier diagnoses to save patients. Compared to other existing methods, this proposed system achieves higher performance. The disease prediction accuracy reached an impressive 96.5%, the precision rate was up to 95.9%, the recall was 96.1%, and the F-measure was 96.5 with redundant time complexity. Identification supports early disease prediction for diagnosis and treatment recommendations. © 2025 selection and editorial matter, Diptendu Sinha Roy, Mir Wajahat Hussain, K. Hemant Kumar Reddy, Deepak Gupta; individual chapters, the contributors. |
