Internet of Things (IoT) is a significant area in the digital era for the purpose of data collection and transferring them on the network without the help of a human, which shows that the whole world is linked over a single system. It can also be seen as the powerful force that operates modern health systems, home automation, improved manufacturing, and smart cities. IoT increases the chances of cyber threats due to the extensive usage of IoT devices and services. Thus, there is a required to design a robust Intrusion Detection System (IDS) to obtain better network security. The conventional machine learning methods are not optimum for processing complex network data by various intrusion methods. Subsequently, the conventional deep learning approaches in intrusion detection show their efficiency in only one-dimensional feature data, and also they are insufficient for predicting the unknown intrusions. This paper focuses on proposing a novel High Ranking-based Optimized Ensemble Learning Model (HR-OELM) using three different classifiers for developing an intelligent IDS. The first phase is data collection, in which the benchmark datasets are gathered. As the features or attributes associated with the IoT devices from benchmark source datasets are more, it is required to extract the most relevant data that could be highly efficient for attaining the high detection rate. Thus, the accurate feature selection is developed to construct a powerful classification methgods and to decreases the data dimensionality. The major highlight of the optimal feature selection is to decreases the correlation between the features giving unique information. These features are subjected to the proposed HR-OELM, in which the Deep Neural Network (DNN), Random Forest, and Adaboost classifiers are used. The detection performance is finalized based on the high ranking of output from three classifiers. One of the main contributions of the proposed IDS is the development of Adaptive Frequency-based Electric Fish Optimization (AF-EFO) for the optimal feature selection and variable optimization of HR-OELM, thus ensuring superior performance. Finally, the suggested ensemble learning model holds a minimum false-positive range and a maximum detection range than the other conventional traditional methods.