Malicious URL Detection Based on Covariance Coefficient and Feature Fusion

Currently, malicious URL detection faces the challenge of dealing with the variable and complex nature of the dataset, where the model may suffer from bottlenecks in feature extraction from the raw data, and thus appears to be not sharp enough in detecting specific anomalous traffic. To address such problems, a malicious URL detection model based on covariance coefficient and feature fusion is proposed. The model avoids manual intervention by generating the covariance coefficient matrix of the raw data and connecting the feature encoding of the raw data with the covariance coefficient matrix through a fully connected neural network. The model extracts multiple effective features from the original data and reliably distinguishes abnormal and normal traffic through validation and correction. Experimental results show that the model improves the accuracy of the supervised model on several datasets and the experimental approach allows for the learning of more generalized models: for example, in the approach using only a normal CNN, the model's accuracy improves by up to 0.2%.