Operating at an ultra-lean air/gas ratio, a modern dry low-emission gas turbine can achieve extremely low emissions levels and meet the strict environmental regulation on NOx emissions. On the other hand, ultra-lean combustion is highly prone to tripping caused by thermoacoustic combustion instabilities and Lean Blowout (LBO), which can result in large pressure oscillations in the combustor and decreased component durability. The prediction methods employed for the early identification of an LBO event prior to its occurrence provide an innovative approach to handling LBO. Prior research on LBO prediction focused mostly on laboratory-scale combustion utilizing flame characteristics as the only parameter, without associating the whole gas turbine, which was a poor and inadequate model of the industrial gas turbine. However, the data-driven model based on the historical datasets has shown greater accuracy and replicativity and has not thoroughly explored LBO prediction. This study investigates the LBO error using an entirely data-driven technique using a dataset collected from an actual ultra-lean dry low emission (DLE) gas turbine’s log file and applying the decision tree classifier for LBO prediction in comparison to logistic regression. The study shows that the LBO can be detected at least one minute prior to the trip. Furthermore, the decision tree classifier model for LBO prediction shows a robust performance with 99.8% accuracy and outperforms the logistic regression model.