A novel approach for sand liquefaction prediction via local mean-based pseudo nearest neighbor algorithm and its engineering application

Abstract The prediction method plays crucial roles in accurate prediction of sand liquefaction. Recently, machine learning has been widely used for prediction of sand liquefaction, and the Local Mean-based Pseudo Nearest Neighbor (LMPNN) algorithm, one of machine learning techniques, showed good performance in pattern recognition. In this study, we propose a sand liquefaction prediction model based on the LMPNN algorithm, which is the first work of applying the LMPNN algorithm to sand liquefaction prediction. Then, our proposed prediction model is used for evaluation of site liquefaction grade in Tongzhou District of China. And the comparison between our proposed prediction model with the liquefaction evaluation method in the Chinese code is made, which will provide an important approach to predicting the sand liquefaction grades for the major construction project sites. Extensive experiments on grade prediction demonstrate that the effectiveness of our proposed prediction model based on the LMPNN algorithm. In addition, shaking table test of an engineering site model is conducted for evaluating whether this engineering site model is liquefaction and non-liquefaction or not. And the experiment result of the shaking table test is the same as that of our proposed prediction model based on LMPNN algorithm, which further demonstrates the effectiveness of our proposed prediction model. Consequently, our proposed prediction model is proved to have a good prospect of engineering application in the liquefaction prediction.