Robust anomaly detection in IoT healthcare using ML with SMOTE and real-Time optimization
Dhirendra Kumar Shukla, Ujjal Kumar Das, Akhilesh Kumar, Arvind Dagur, Shabir Ali · 2025
The integration of IoT technologies in healthcare enhances monitoring and diagnostics but introduces significant security risks. This study proposes a robust anomaly detection framework utilizing the CICIoT2023 dataset, addressing class imbalance and redundant features through SMOTE and Pearson correlation. Multiple machine learning models—including Random Forest, Adaptive Boosting, and Deep Neural Networks—are evaluated. The proposed method achieves outstanding accuracy: 99.68% (2-Class), 99.44% (8-Class), and 99.16% (34-Class), outperforming all baseline models. These results demonstrate the framework&s;s effectiveness in detecting complex cyber threats. Its low-latency design supports real-time inference, making it well-suited for deployment in IoT-based healthcare environments.