The classical Tikhonov regularization has been successfully invoked in supervised learning and proved to be consistent. Motivated from the works of De Vito et al. (J Mach Learn Res 6(5) 2005), we introduce a generalization of the Tikhonov scheme, called weighted-II algorithm and investigate its consistency. A parameter choice rule based on the Hölder source condition has been derived in order to establish the consistency and a discussion of rate of convergence is also pursued.