Recent functional magnetic resonance imaging (fMRI) studies have identified specific neural patterns related to three different categories of movements: intransitive (i.e., meaningful gestures that do not include the use of objects), transitive (i.e., actions involving an object), and tool-mediated (i.e., actions involving a tool to interact with an object). However, fMRI intrinsically limits the exploitation of these results in a real scenario, such as a brain-machine interface. In this paper, we propose a new approach to automatically predict intransitive, transitive, or tool-mediated movements of the upper limb using electroencephalography (EEG) spectra estimated during a motor planning phase. To this end, high-resolution EEG data gathered from 33 healthy subjects were used as input of a three-class k-nearest neighbors classifier. Different combinations of EEG-derived spatial and frequency information were investigated to find the most accurate feature vector. In addition, we studied gender differences further splitting the dataset into only-male data, and only-female data. A remarkable difference was found between accuracies achieved with male and female data, the latter yielding the best performance (78.55% of accuracy for the prediction of intransitive, transitive, and tool-mediated actions). These results potentially suggest that different gender-based models should be employed for the future BMI applications.