Leak quantification requires accurate data from sensors to achieve high localization accuracy. This task requires sensors to continuously send data to a central which consumes significant sensors' energy. We aim to preserve sensors' energy by allowing only sensors with significant readings to send their data to the central and preventing sensors with insignificant readings to do so. The challenge is to determine which sensor whose data is important to maintain the accuracy of leak quantification. We propose a method to classify the importance of sensor data based on the separation of leak signatures that it provides. Then we set a threshold that a sensor has to meet before sending data to the central. The threshold is set based on the sensor's classification. The more important the sensor, the lower the threshold, and thus, the more frequent data the sensor sends. Our experimental results have shown that our method can maintain a decent localization accuracy indicated by 6% to 9% decreases from the highest accuracy achieved by using all sensor readings (90%). This result indicates that our feature selection method could be used to preserve sensor energy in wireless sensor networks.