The space-weather complexes including characteristics of solar streams (parameters of the solar wind, components of the interplanetary magnetic-field vector) and temporal quantitative estimates of their geomagnetic effects (Dst index) are classified. Comparative and neural-network methods for this classification are developed. As a result of numerical neural-network experiments, types of solar streams responsible for generation of geomagnetic perturbations with different intensities are established. It is confirmed that, on the basis of the selected space weather complexes, it is possible to refine the influence of these streams on the electromagnetic state of the magnetic sphere and, hence, improve the accuracy of predictions of this state.