The paper investigates the classification of oil temperature ranges during the frying process through the utilization of audio and video modalities, as well as their fusion. Data was collected in a kitchen, using smartphone audio and video sensors. Wet chopsticks were used as a controlled stimulus to cause a bubble production in the hot oil, with temperatures ranging from 140 to 240  \(^\circ \) C in 10  \(^\circ \) C increments. Corresponding data were recorded at each temperature interval. Pre-processing of the collected audio and video data was performed to eliminate irrelevant information, facilitating the training of a deep learning model. The findings highlight the significance of bubbles, including their number and size, as main features for classification, with a more significant presence of numerous and more giant bubbles observed at higher temperatures. Ultimately, the study demonstrates that the accuracy of audio and video-only modalities can be enhanced by applying multimodal learning techniques.