One of the most important aspects of machine fault prognosis is the selection of sensors and features to represent the degradation process of a faulty component. Several approaches in the literature have used features extracted from vibration signals to estimate the future condition based on time series forecasting. Another technology that has been used increasingly for this task are the acoustic emission (AE) sensors, which have frequency measurement ranges much higher than vibration sensors. On gearboxes some studies have shown that the AE technology can be used effectively for fault diagnosis, but its use for fault prognosis is still a relative new field of research that offers encouraging opportunities. One downside of the application of the AE technology in gearboxes is the strong dependence of the AE on the oil temperature, which may lead to difficulties during the forecasting of an AE-based feature. Thus, in this study a novel feature based on a relative counting of the AE bursts is proposed and tested with data from a planetary gearbox with a ring gear fault. The proposed feature reduces the influence of the temperature on the generation of AE when it is compared to the counting based on a fixed amplitude threshold. Therefore, it can then be more suitable for fault prognosis than traditional AE counting. In this case study the forecasting of the proposed feature is carried out using an artificial neural network (ANN), whose hyperparameters were selected using a genetic algorithm. The results are promising and constitute a basis for further research.