How to effectively respond to tax risk identification and management challenges amid the proliferation of big data is an urgent problem for enterprises. In this study, the data sources are firstly clarified, and the redundancy and noise are eliminated through preprocessing, which ensures the data quality and enhances the credibility of the subsequent conclusions. The features of enterprise tax risk management are imported into SVM as inputs for identification and prediction, and the traditional SVM algorithm takes a long time to compute when facing a large amount of feature data. To address this issue, the least squares (LS) method is introduced to optimize the conventional SVM framework, converting the quadratic programming challenge into a system of linear equations, thereby maximizing the identification performance, and finally obtain a LS-SVM-based tax risk management identification algorithm. -SVM-based recognition model for tax risk management. With the theoretical support of research dataset, confusion matrix, and quantitative assessment indexes, the model is deeply explored and analyzed. The LS-SVM hybrid algorithm achieves an accuracy of 0.9772, precision of 0.9778, recall of 0.9911, F1 score of 0.9844, and AUC of 0.9922, all of which surpass the performance of the decision tree, random forest, and logistic regression approaches in recognition, interpreting the application value of LS-SVM algorithm in tax risk identification and management.
