<p>Support Vector Machine (SVM) is a valuable tool for binary classification, but its performance can drop significantly if irrelevant variables are included, making variable selection critical. With the rise of big data, we often encounter streaming data that comes in batches and has no fixed size. This makes traditional SVM variable selection impractical, and how to update without saving the raw data is the main issue. Also, the hinge loss function’s non-smoothness poses challenges in computation, especially for the large-scale problems. To solve these problems, we propose an online updating SVM variable selection procedure, which involves only the current data and summary information of historical data. A key benefit of this method is its ease of implementation and low memory demands. Theoretically, this online updating method is asymptotically equivalent to the standard version computed on entire data. Numerical experiments confirm that it performs well.</p>