Abstract: Aiming at  the problems of missing data and small number of features in landslide disaster prediction around natural gas pipelines, 
the gradient boosting decision tree algorithm based on LightGBM framework was adopted to  supplement missing data by interpolation,  and short-term and long-term features were  generated by using the historical feature data to obtain the importance ranking of various factors affecting the slope evolution process and the optimal parameter set of the algorithm, so as to realize the effective prediction of landslide disasters around natural gas pipelines. The results show that  this method has higher accuracy and faster processing speed than XGBoost model  in the prediction of landslide disasters around the pipeline, which proves that LightGBM algorithm is feasible and effective in landslide disaster prediction.

Key words: LightGBM framework, landslide, early-warning and forecast, machine learning, artificial intelligence