In a network traffic, detecting anomalies is currently considered an ongoing and critical research area, particularly when Internet of Thing devices (IoT) are connected. IoT devices are quickly gaining importance in the present scenario in every nook and corner of people’s lives, and they are also highly prone to attacks from various malicious devices. The proposed paper introduces a technique for handling IoT devices’ evolving anomaly detection issue. It is done by concatenating Kaggle datasets for DoS attacks obtained from anomalous IoT traffic and assessing them using an ML model that can identify malicious and normal IoT traffic and various anomaly types. A massive dataset with data approaching from several IoT scenarios is combined to provide an accurate and still missing benchmark for abnormal and normal IoT traffic. Furthermore, the ML model has been enhanced with a feature reduction phase by utilizing the learning model and a robustness study of the best-weighted sum feature analysis (WSFA) in conditions affected by external noise. The IoT dataset developed gained the best outcomes, and its effectiveness is proven for anomaly detection through ML techniques, even in scenarios of unwanted noise.