A computer implemented method predicts by a neural network a radio spectrum usage in an environment comprising interferers and a plurality of wireless nodes. The nodes operate via a slotted medium access control, MAC, protocol within a radio spectrum. The method includes the following steps: obtaining per slot the radio spectrum usage by sensing the radio spectrum, labelling the slots as usable when a respective radio spectrum usage is lower than a predefined noise threshold or when a node successfully transmitted a packet through it, unusable when it is not usable and suitable to receive a packet, and calculating a loss function as a cross entropy loss with the usable and unusable slots over a total number of usable and unusable slots, and training the neural network by minimizing the loss function.