The problem of decreased balance caused by injury, illness, or aging is becoming increasingly prevalent in society. The traditional functional balance assessment method is highly time-consuming and inefficient, as well as being susceptible to measurement errors caused by personal subjective factors. This study proposes a system that can rapidly, conveniently, and accurately predict a participant's Berg balance scale (BBS) score without professional supervision. The proposed system uses a wearable inertial sensing device combined with machine learning to predict the BBS score of a test participant. In the beginning, the participants were asked to wear inertial sensing devices on seven parts of the body and perform 17 test tasks. The wearable device locations and the test tasks were ranked by importance and further reduced wearable devices and the test tasks. Eventually, the participant is only required to wear an inertial sensing device on their left thigh and perform two simple test tasks, namely "placing an alternate foot on a stool" and "standing on one foot (right foot)," to obtain their BBS score. In this study, the proposed system has a high level of accuracy for predicting BBS scores. The experimental results indicate that the mean absolute error (MAE) of the proposed system was 1.274. Moreover, this study provided some important information as a reference for future research on functional balance, including feature sets selection, regression model selection, wearable device locations ranking, and test tasks ranking. The researchers can use that information to design their experiment.