Abstract:

The data mining techniques are used to predict loess collapsibility in geotechnical engineering; and the mining model is constructed by using the least squares support vector machines. Using principal component analysis method, data of model is preprocessed to remove the correlation among the indicators and to eliminate the impact of multi-index redundant information on the mining model, and the model inverse analysis is administered by introducing particle swarm optimization algorithm to determine the optimal parameters. The forecast mining for the actual project data shows that loess resistivity and shear wave velocity is closely related to the soil indicators such as the soil structural properties, water content, density etc, they can be more comprehensively reflect the factors of impact of loess collapsibility. Using loess resistivity, shear wave velocity and soil depth as predicting variables of the model can quantitative predict the loess collapsibility; the proposed model is effective.

Key words: loess, collapsibility prediction, data mining, least squares support vector machines