Abstract:
Cloud-based healthcare networks generate vast amounts of data, making anomaly detection
essential for ensuring data integrity and system reliability. However, existing methods struggle
with high false positive rates, scalability limitations, and increased detection latency, reducing
their effectiveness in healthcare systems. To overcome these challenges, this paper introduces a
robust anomaly detection framework for cloud-based healthcare networks using deep learning
techniques. The framework begins by collecting data from healthcare environments, the collected
data is then subjected to data preprocessing, where KNN imputation is applied to handle missing
values, ensuring completeness, while Winsorization mitigates extreme outliers to maintain data
consistency and improve reliability. The preprocessed data is then used for feature extraction,
where Polynomial Feature Expansion is applied to generate higher-order features, capturing
complex relationships between healthcare parameters. The extracted features are then fed into a
Temporal Convolutional Network, which learns temporal dependencies within sequential
healthcare data, effectively distinguishing between normal and anomalous patterns with high
accuracy. Once anomalies are detected, they are stored in cloud storage, ensuring efficient
management and seamless integration for further analysis. Experimental results validate the
proposed framework, achieving 99.52% accuracy, 98.34% precision, 98.78% sensitivity, 97.23%
specificity, and 99.74% F1-score, significantly outperforming conventional approaches. This work
enhances the computational efficiency, reducing false positives, and improving detection
reliability, making it a scalable and robust solution for cloud-based healthcare applications. |
