Centrifugal compressors used in floating production storage and offloading (FPSO) have special characteristics, forcing manufacturers to build machines that operate at higher speeds in order to achieve the required process performance. Under these operating and design conditions, a special balance procedure called high-speed balancing may be necessary. This procedure is generally done in a special workbench of specialized companies and carrying out a field balance of these machines is an unlikely alternative. Therefore, the development of techniques to identify unbalance malfunction prior to reaching the machine vibration trip limits is essential to minimize the unit downtime. Data-based techniques are strong candidates for the task, since they can take full advantage of huge amounts of available data, allowing the information to be used in an effective way. In this work, two different techniques were chosen to improve unbalance identification: Mahalanobis–Taguchi system (MTS) and principal component analysis (PCA). Considering a real machine, several unbalance conditions were simulated and the results are presented and discussed. Both methods were able to detect unbalance, with a slightly advantage for PCA for the considered conditions.