The rapid proliferation of IoT devices has been raising concerns on the dynamic and heterogeneous networks for security, resource efficiency, and scalability. Most methods applied in IoT security often face trade-offs between robustness, computational efficiency, and adaptability. Conventional authentication techniques do not have the flexibility to address uncertain and dynamic conditions while centralized logging systems are susceptible to tampering and single points of failure. IDSs also fail to detect zero-day attacks most of the time, and protocol parameters are rarely optimized for diverse IoT environments. We propose a multi-faceted framework that integrates Fuzzy Logic-Based Adaptive Authentication (FLAA), Blockchain for Decentralized Secure Logging (BDSL), Deep Learning-Based Intrusion Detection (DLID), and Metaheuristic Optimization of Protocol Parameters (MOPP). FLAA authenticates devices dynamically using fuzzy rules that adapt to trust scores, signal strength, and reliability metrics, reducing authentication delay by 30% and maintaining a false-positive rate below 5%. BDSL uses lightweight Proof-of-Stake consensus mechanism to ensure tamper-proof logging of authentication events with less than 5% network overhead. DLID uses a Transformer-based model that achieves real-timestamp anomaly detection. It achieved 98.5% accuracy on known attacks and 92% on zero-day attacks. Finally, MOPP uses Genetic Algorithms optimizing security, latency, as well as energy consumption and reduces energy usage by up to 25% and latency by up to 20%. This approach hybridizes the IoT security of kernels using adaptive, transparent, scalable solutions while using minimal amounts of resources. Our framework, using adaptive authentication, immutable logging, and intelligent intrusion detection, promises robust security with operational efficiency in conjunction with the different IoT ecosystems.