The purpose of this study was to investigate whether or not a computer may predict the outcome of soundscape assessments, based on acoustic data only. It may be argued that this is impossible, because a computer lack life experience. Moreover, if the computer was able to make an accurate prediction, we also wanted to know what information it needed to make this prediction. We recruited 33 students (18 female; Mage = 25.4 yrs., SDage = 3.6) out of which 30 assessed how pleasant and eventful 102 unique soundscape excerpts (30 s) from Stockholm were. Based on the Bag of Frames approach, a Support Vector Regression learning algorithm was used to identify relationships between various acoustic features of the acoustics signals and perceived affective quality. We found that the Mel-Frequency Cepstral Coefficients provided strong predictions for both Pleasantness (R2 = 0.74) and Eventfulness (R2 = 0.83). This model performed better than the average individual in the experiment in terms of internal consistency of individual assessments. Taken together, the results show that a computer can predict the outcome of soundscape assessments, which is promising for future soundscape mapping.