Under the rapid progress of Internet technology, phishing attacks have become a serious threat in network security. However, traditional decision tree algorithms often encounter the dual difficulties of unstable classification accuracy and low computational efficiency in recognizing such attacks. To address this problem, this paper focuses on creating an improved C4.5 decision tree algorithm that integrates the boundary point principle with learning vector quantization. By virtue of the boundary point principle, this algorithm effectively reduces the number of candidate segmentation points and greatly improves the efficiency of the algorithm. On top of that, the learning vector quantization approach is introduced to intelligently cluster the raw data, which in turn optimizes the segmentation point selection mechanism. More importantly, by deeply integrating the information entropy and covariance characteristics together, a set of attribute selection mechanism with higher accuracy is constructed. The results show that the proposed algorithm exhibits excellent classification performance when dealing with a wide range of datasets. Especially when dealing with high-dimensional and complex data environments, not only the classification accuracy is significantly improved, but also the computational efficiency has an essential leap. This study provides an efficient and accurate solution for phishing attack identification, which not only has the value of theoretical research, but also presents a broad application prospect and far-reaching social significance in practical application.
