The author proposes a safe screening rule with bi-level optimization for ν-SVM to reduce computational cost without sacrificing prediction accuracy.