<p>Cyberattacks provide significant security risks for modern internet and cloud-based technology. The rapid advancement of Internet of Things (IoT) technology has led to the emergence of security concerns in organizations, including botnet attacks, malware, ransomware, and encryption. These potential hazards may pose challenges because of their vulnerability to hacking and ability to self-configure. Malicious attackers launch attacks by exploiting multiple security vulnerabilities in IoT networks. Due to the distinctive network designs, limited resources of nodes, and network protocols of the IoT, current multilabel attack detection techniques are inadequate for IoT systems. Recently, deep learning demonstrated its efficacy in precisely recognizing malware activities from network traffic data. The study developed an autoencoder-based Multi-Label Attack Detection Algorithm (MADA) to identify attacks in the IoT framework. The study used three crucial datasets, such as UNSW-NB 15 IoT, DDoS-attack, and Cyberattacks RT-IoT2022, which contain multiple security malicious attacks. The datasets consist of standard instances and attack threats, with a dataset containing million records belonging to a classified category of attack, ensuring comprehensive protection against potential threats. The experimental findings show that the proposed framework has a superior capacity to identify and mitigate attacks compared to existing approaches. The proposed final framework attained an accuracy of 99.96% during training and 99.39% during testing, with a testing recall of 99.43% and an F1 measure of 99.38%. The validation loss of the proposed framework was the lowest at 0.0153. We compared the computational complexity and performance before and after applying SMOTE. We also compared the proposed model with efficient base-line models and performed statistical tests to validate the model’s effectiveness. Additionally, the proposed framework is more efficient, fast, and accurate to detect attacks over IoT networks.</p>