As the demand for wind power continues to grow at an exponential rate, reducing operation and maintenance expenses and improving reliability has become pinnacle priorities in wind turbine maintenance strategies. Prediction of wind turbine failure earlier than they reach a catastrophic degree is essential to reduce the operation and maintenance cost because of unnecessary scheduled maintenance. In this study, a SCADA-data based condition monitoring system is proposed using machine learning techniques. We trained various machine learning models using our dataset, and then selected the best among those to predict the gearbox temperature. The bagging regression method accomplished the best accuracy with 99.7% R2 score, while restraining the mean square error to 0.35. The experimental results showed that our method anticipated 68 days ahead of turbine gearbox failure, and generated another alarm when fault turned intense. The time between alarms and actual failure is enough for the operator to fix the gearbox before it turns to a catastrophic event.