With the advent of wireless communication and digital technology, low power, Internet-enabled, and reconfigurable wireless devices have been developed, which revolutionized day-to-day human life and the economy across the globe. These devices are realized by leveraging the features of sensing, processing the data and nodes communications. The scale of Internet-enabled wireless devices has increased daily, and these devices are exposed to various cyberattacks. Since the complexity and dynamics of the attacks on the devices are computationally high, intelligent, scalable and high-speed intrusion detection systems (IDS) are required. Moreover, the wireless devices are battery-driven; implementing them would consume more energy, weakening the accuracy of detecting the attacks. Hence the design of the IDS is required, which has to establish the good trade-offs between Energy and accuracy. This research includes the Multi-tiered Intrusion Detection (MDIT) with hybrid deep learning models for improved detection accuracy in wireless networks; spotted hyena optimization (SHO) and Long short-term memory (LSTM) have been studied to design IDS effectively. Extensive experimentation has been carried out in real-time scenarios using the Node MCU Embedded boards and standard benchmarks such as CIDDS-001, UNSWNB15 and KDD++ datasets compared with the other traditional and existing learning models. The average prediction accuracy of 99.89% for all datasets has been achieved. The results show that the proposed system guarantees a high detection accuracy and reduces the prediction time, making this system suitable for resource-constrained IP-enabled wireless devices.