Landslide displacement prediction based on spatio-temporal association rule mining between target case and similar cases

The displacement response of landslides to external influencing factors varies according to different creep stages. However, most traditional models ignored this variation and yielded low prediction accuracy especially for mutational displacements of acceleration deformation. To overcome this drawback, this study utilized the spatio-temporal association between the similar landslide cases with sufficient monitoring data and target case to replace the correlation between landslide displacement and external influencing factors to improve the prediction capacity for the mutational displacements. In order to reduce the prediction error caused by the difference between landslide cases, weights were assigned for similar cases according to the similarity of geological and hydrological conditions between cases in realizing the final weighted average prediction for target case. The model barely needs the long-time monitoring of external influencing factors, and it can not only enrich the data through historical landslide cases, but also reduce monitoring cost for target case. The findings of this study demonstrated the presence of a novel perspective for prediction, which may reduce prediction errors caused by any significant deviations in landslide from its previous deformation behavior. Application in engineering, as well as comparing prediction results from different models, may highlight the advantages of the proposed method.