Supervised Anomaly Detection for Features Extraction from Normal Semantics

In the realm of anomaly detection, identifying deviations from normal behavior in data is critical across various applications, including cybersecurity, manufacturing, and healthcare. Traditional supervised methods rely on labeled datasets, which are often scarce or expensive to obtain. To address this challenge, we propose supervised anomaly detection framework that leverages deep learning techniques to extract features from normal semantic patterns. Our approach integrates Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and ResNet-18 architectures to enhance the detection accuracy and efficiency. Initially, CNNs are employed to capture spatial hierarchies in the data, learning robust feature representations from normal samples. The extracted features are then fine-tuned using the ResNet-18 model, known for its depth and skip connections, to ensure comprehensive feature extraction and minimize information loss. Finally, these deep features are fed into an SVM to differentiate between normal and anomalous instances based on the learned semantics. Experimental results on benchmark datasets demonstrate the superiority of our method in detecting anomalies with high precision and recall, outperforming traditional anomaly detection techniques. Our framework’s ability to operate without labeled anomalies and adapt to various domains underscores its potential for broad applicability in real-world scenarios.