Microbubble emission boiling (MEB) is a cooling technology in which the heat flux can potentially exceed the critical heat flux (CHF). Reliable predictions of the occurrence of MEB are necessary to achieve stable MEB and to induce it under actual environment conditions. In this study, we developed a method based on deep learning with boiling sound to predict the boiling state of the interval before the low-heat-flux level reaches MEB. The boiling sound was acquired by a hydrophone, and the sound was adopted to machine learning algorithms, which were subsequently applied to classification and regression models. The feature extraction algorithms for the boiling sounds were spectrum or cepstrum methods. Both methods were comparatively investigated in terms of the machine learning accuracy. As a result, in the case of the cepstrum method as the feature extraction, the accuracy was improved. In particular, we found that the regression model demonstrated substantially better accuracy than the classification model. In addition, accurate predictions were possible even when the degree of subcooling was changed.