Numerous of IoT devices are in fact vulnerable to cyber threats. These vulnerabilities may be exploited over the internet and by remote access by the malefactors. Motivated by this, we suggest a reliable, knowledgeable IoT networks threat detection system. In this research, it aims at evaluating the integration of dual key design concepts in the development of a deep learning based intelligent threat detection system located at the edge of the IoT network. Given these notions, we introduce the ShieldNetMapper (SNM-Rand-N-MMOA) model. Finally, real time IoT traffic data is pre-processed using Spark, Enhanced Random Neural Network (Rand-NN) and Multifaceted Mayfly Optimization Algorithm (MMOA). A deep model is extracted by MMOA from the range of significant weights and bias values for the purpose of reconstructing the optimal network traffic data. On the other hand, we are utilizing Rand-NN for classification as well as to avoid the deep learning model from overfitting. This hereby proposes the model that evaluates IoT real time dataset using the recall rate of 99.78 % and average accuracy is 99.81%. With these impressive results, this is proof that the model can actually separate between different types of IoT traffic and this can help enhance the security and efficiency of operations in the network. Therefore, by integrating MMOA and Rand-NN, not only are data processed in an optimal manner but also the robust classification in dynamic environments is also guaranteed.