The linking of abnormal and malicious business in the network is crucial for IoT safety to monitor and restrict unauthorised IoT network traffic flows. Many machine literacy (ML) approach models have been provided over by various experimenters to aid in preventing malicious business movements in the IoT network. Due to the sloppy point selection, several ML models are vulnerable to misclassifying potentially harmful business flows. However, additional research has to be done to fully understand how to pick appropriate features for the accurate identification of malicious enterprises in IoT networks. To address the issue, a new frame model is proposed. First, we propose a new point selection measure entitled CorrAUC, and then, based on CorrAUC, we create and design a new point selection method called Corrauc. Corrauc uses a wrapper-style technique to evaluate the features directly and select the most effective features for the specified ML algorithm. In addition, we used a combined TOPSIS and bijective soft set to verify the names of features used to identify malicious enterprises in the IoT system. We evaluate the effectiveness of our suggested method using the Bot- IoT dataset and four distinct ML methods. The examination of experimental data confirmed that our projected system is efficient and can produce > 96 consequences on average.