The application of the support vector machine (SVM) classification algorithm to large-scale datasets is limited due to its use of a large number of support vectors and dependency of its performance on its kernel parameter. In this paper, SVM is redefined as a control system and iterative learning control (ILC) method is used to optimize SVM’s kernel parameter. The ILC technique first defines an error equation and then iteratively updates the kernel function and its regularization parameter using the training error and the previous state of the system. The closed loop structure of the proposed algorithm increases the robustness of the technique to uncertainty and improves its convergence speed. Experimental results were generated using nine standard benchmark datasets covering a wide range of applications. Experimental results show that the proposed method generates superior or very competitive results in term of accuracy than those of classical and state-of-the-art SVM based techniques while using a significantly smaller number of support vectors.