The concept of the Internet of Things (IoT) has becomes very popular and leads to significant increase in the number of IoT networks, appliances, processed data. For such IoT appliances, security is becoming a top priority since they create and share sensitive data via the conventional Internet. In prior researches, for safeguarding the network entry points and continuous monitoring of network traffic, a deep neural network (DNN)-based intrusion detection system (IDS) is used which provides effective security to the IoT but with a high false alarm rate (FAR). So, this research presents an improved dwarf mongoose optimization (IDMO) algorithm for tuning the parameters of DNN for effectively identifying the network anomalies with reducing FAR. DNN parameters are fine-tuned using IDMO for improving the anomaly detection. Finally, using the input features, the DNN detects the anomaly. When compared with the conventional techniques such as recursive feature elimination with DNN (RFE-DNN), two convolutional neural networks (CNN-CNN), and DNN, the IDMO-DNN performs better through achieving the higher detection accuracy of 99.97%.