<p>Wireless Sensor Networks integrated with Internet of Things (WSN-IoT) are exceedingly vulnerable in regard to security based on the significant amount of confidential data that is processed by them and device interconnectivity between heterogeneous applications such as smart cities and healthcare. The solutions must be low-latency, accurate, and scalable for preserving the confidentiality and integrity of the data and supporting the system performance. In fact, such systems are suffering from serious challenges resulting from different types of cyber threats because of intrusion attempts that may lead to data breaches and network operations. This is because most of the current models represent intrusive detection systems and thus suffer in terms of accuracy and efficiency, as their high-dimensional datasets with traditional classification methods may increase false positives and slow down the response. The two main novelties presented in this paper are AqRO for feature selection techniques and CapTAD for intrusion detection. On the other hand, the CapTAD model uses capsule networks to learn the temporal dynamics from network traffic, achieving very high classification accuracy for various intrusion types with robustness to noise and variability in the input data. The novelty of the work is the synergistic integration of AqRO and CapTAD, which facilitates not only streamlining in feature selection but also enhances classification performance in cloud-WSN environments. It results in huge improvement in accuracy and precision of the detection and overall efficiency of the system by combining these techniques, hence improving the limitations of previous models. We further validate our methods against benchmark datasets such as WSN-DS and UNSW-NB 15 for their effectiveness and robustness in the identification and mitigation of cyber threats. This research enriches the ongoing efforts toward securing Cloud-WSNs and provides a base for their future improvements on intrusion detection systems.</p>