Abstract: Pipeline transportation is the most economical way to transport oil, natural gas, and other energy over a long distance. Excellent low-temperature toughness is one of the important characteristics to ensure the safe transportation of pipeline steel. Drop weight tear testing （DWTT） is the most effective method to measure the low-temperature toughness of pipeline steel. In the present work, a DWTT prediction model based on machine learning is established according to the production line data set provided by the steel mill and the literature data-assisted production line data set. Based on two data sets, different machine learning algorithms are tested. The best models are random forest models. The accuracy of Strategy I with pure production line data is 0.64, and the accuracy of Strategy II with literature data-assisted production line data is 0.92, wherein literature data effectively help improve the DWTT prediction accuracy. The machine learning model provides a new method for optimizing and predicting DWTT.