IoT is defined as the network of devices that frequently communicate to exchange sensitive data including control signals, environmental sensor readings, and health records. This exchange is vulnerable to various attacks that primarily try to disturb the normal working of network by denying legitimate services and stealing information. Therefore, timely detection and mitigation of these attacks are important to make IoT services acceptable. In this study, AI based detection system for four types of attacks: DDoS, DoS, OS fingerprinting and theft is proposed. The proposed system uses a novel feature set for detection of each type of attack. The feature set is selected based on Gini Importance. In addition to this a new feature ETV is also proposed. Further, experimental results demonstrate that the model trained on this combined feature set achieves high accuracy, high precision while maintaining a low False Alarm Rate (FAR). Further, a comparative analysis of supervised machine learning algorithms including Support Vector Machine, Logistic Regression, Random Forest, and Decision Tree was conducted to detect four distinct types of attacks. Additionally, performance of each classifier is validated on anomaly detection. Among these, the Random Forest classifier outperform the others, achieving a 0.3% FAR, 99.8% accuracy, and 99.9% precision.