Date of Publication :3rd June 2024

Abstract:The framework that uses sophisticated deep learning algorithms and data preparation to improve anomaly detection in Internet of Things systems. In order to standardize location coordinates and minimize the computational load on the network, the first data transformation uses Min-Max normalization. Next, it is shown how to use Principal Component Analysis (PCA) to efficiently reduce dimensionality while maintaining important information in high-dimensional datasets that are frequently used in Internet of Things applications. The process of standardization in principle component analysis (PCA) guarantees fair feature contributions. The covariance matrix is then computed, which makes it easier to extract principal components and capture the maximum variance in the data. Additionally, by using CNNs' ability to autonomously learn hierarchical representations straight from pictures, the paper suggests integrating CNNs for image-based anomaly identification. The CNNs are very good at identifying abnormal from normal patterns across a wide range of domains because they use transfer learning and encoder-decoder architectures to capture complex patterns. With accuracy of 90.91%, recall of 87.9%, F1-score of 90.3%, and a ROC value of 95%, the proposed CNN model shows encouraging results, highlighting its resilience in anomaly identification. Looking ahead, the area of work includes improving methods for detecting anomalies through creative pretreatment of data and fine-tuning CNN structures to make them more flexible in the face of changing Internet of Things scenarios. The investigation of ensemble methods and reinforcement learning offers further opportunities to boost anomaly detection systems' accuracy and robustness. Overall, this study offers a thorough and practical method for IoT anomaly detection, adding to the changing field of intelligent and connected devices.

Reference :