Abstract: Focusing on non-linear separable network data with unknown specification, an unsupervised feature selection algorithm based on Support Vector Machine (SVM) was proposed, termed UFSSVM. The proposed algorithm first maps the non-linear network data into a high dimensional feature space using a non-linear mapping function; then it performs unsupervised feature selection in the high dimensional feature space. Compared with traditional unsupervised feature selection algorithms, the proposed algorithm can automatically get the relevant features just using the original network packet without the preprocessing step to get the original feature set. The performance of the proposed algorithm is examined by simulations and with real network data set. Experiment results illustrate the feasibility and effectiveness of the proposed algorithm in feature subset selection.