ISSN 2277-2685 IJESR/October-December 2024/ Volume-14/Issue-4/162-183 Surendar Rama Sitaraman et. al., / International Journal of Engineering & Science Research 162 BI-DIRECTIONAL LSTM WITH REGRESSIVE DROPOUT AND GENERIC FUZZY LOGIC ALONG WITH FEDERATED LEARNING AND EDGE AI-ENABLED IOHT FOR PREDICTING CHRONIC KIDNEY DISEASE Surendar Rama Sitaraman, Intel Corporation, California, USA surendar.rama.sitaraman@ieee.org Poovendran Alagarsundaram, Humetis Technologies Inc, Kingston, NJ, USA poovendrana@ieee.org Harikumar Nagarajan, Global Data Mart Inc (GDM), New Jersey, USA harikumarnagarajan@ieee.org Venkata Surya Bhavana Harish Gollavilli, Under Armour, Maryland, USA venkataharish@ieee.org Kalyan Gattupalli, Yash Tek inc, Ontario, Canada kalyangattupalli@ieee.org Dr.S.Jayanthi M.E.,Ph.D., Principal, Tagore Institute of Engineering and Technology Deviyakurichi, Thalaivasal (TK), Salem sjayanthi@ieee.org ABSTRACT: Chronic Kidney Disease (CKD): A growing public health concern Appropriate and timely diagnosis of chronic diseases is important in determining subsequent interventions. In this research, we present a CKD prediction model with FL, edge AI and Bi-LSTM along with Regressive Dropout and GELU activation to boost the performance of CKD prediction. In addition, Generic Fuzzy Logic (G-Fuzzy) increases the accuracy of CKD stage classification. The computational complexity of the proposed model is further accelerated with a Granular Information-based Krill Herd Algorithm (GI-KHA) that performs feature selection. Data Results: Achieves better prediction ability (98.96%) than traditional approaches The approach is scalable, respects patient data privacy and thus applicable in the healthcare settings where real-time diagnosis of CKD could be leveraged. Objective: Creating a CKD detection model by integrating Edge AI with Federated Learning, and improving the accuracy rate, privacy, as well as real-time usage. Methods: This model uses Bi-LSTM with GELU activation, Regressive Dropout, G-Fuzzy Logic for CKD classification and GI-KHA for feature selection. Privacy in a Distributed Data training is achieved by using Federated Learning. Results: The sensitivity, precision and recall were superior to the traditional model with an accuracy of 98.96%. Conclusion: Our system presents a method for early detection of CKD which is efficient, scalable and privacy- preserving, whereby the proposed solution has the potential to provide enhancements in patient care. Keywords: Chronic Kidney Disease (CKD), Federated Learning, Edge AI, Bi-LSTM, GELU, Regressive Dropout, Generic Fuzzy Logic, Krill Herd Algorithm, Healthcare Prediction.