Abstract: A PCA-ELM model for better classifying expansive soil was proposed in this paper by integrating Principal Component Analysis(PCA) and Extreme Learning Machine(ELM). Four classification indexes well reflecting the swell-shrink characteristics of expansive soils, namely liquid limit, plasticity index, content of clay particles smaller than 2 μm, and free swell ratio, were selected for correlation analysis, and two principal components were determined according to accumulated variance contribution rate.Subsequently, 70% of the samples were divided as training set which was taken as the input of extreme learning machine, and 10-fold cross validation was used to optimize model parameters so as to achieve the optimal classification; 30% of the samples were chosen as test set as the input of the optimal model to obtain classification results.The classification model was validated with 32 testing examples from two engineering projects, and results suggest that the classification results agreed well with measured data, with the classification accuracy of training set and test set reaching 94.20% and 79.00%, respectively; moreover, the proposed model is of fast training speed, hence is suitable for the classification and prediction of large-scale data.

Key words: expansive soil, extreme learning machine, principal component analysis, classification model, cross validation