The centrality of ion-ion collisions is an important parameter that defines the initial geometry of the interaction and significantly influences the interpretation of experimental results. Since centrality cannot be measured by direct experiments, various model-dependent methods are used in practice. In SPD (Spin physics detector) experiment located at NICA collider centrality determination is particularly challenging due to low particle multiplicities in small collision systems at low energies, such as Al+Al collisions at √sNN = 10 GeV. In this work, the applicability of traditional centrality determination methods based on charged-particle multiplicities and spectator nucleon energy deposition is investigated under SPD conditions. In addition, the potential for improving centrality determination by using a set of experimental observables and machine-learning techniques based on decision tree classifiers is explored.