Intrusion detection system (IDS) plays a major role in designing a security system to detect the intruders who affect the efficiency of the networks. mobile ad hoc networks (MANETs) and wireless sensor networks (WSNs) do not have a fixed structure, and the Internet of Things (IoT) is an enhanced paradigm including these types of networks. Due to their distributed architecture and limited resource base, security arenas present immense difficulties for such systems, calling for an adaptive IDS. In the previous sections, we introduced a cross-layer IDS that employed the detection with a two-tier structure. To further this, we propose a deep learning-based IDS using convolution neural network (CNNs) for extraction of features and recurrent neural network (RNN) for temporal patterns in the data. These allow for enhanced detection accuracy; moreover, this approach also provides scalability and ability to adapt to new conditions. The results obtained from the evaluation reveal that the detection ratio varies with the different power/node velocity in which we get a detection ratio higher than 99% for high power/node velocity and more than 95% for low power/node velocity and this proves that the deep learning techniques used in the IDS improve the performance ratio than the previous method used.