Abstract: By sorting out the engineering geological data of 400km² area in the main urban area of Xi'an, about 20793 engineering geological drillings are selected to be used in spatial distribution researching of miscellaneous fill and plain fill. Gaussian mixture clustering algorithm in Machine learning is used for learning unlabeled drilling data samples, Akaike Information Criterion and Bayesian Information Criterion are used for testing Gaussian Mixture Clustering Algorithm, and n=140 is the bottom of cluster number of miscellaneous fill and plain fill are determined by trial calculation, and then spatial distribution map of miscellaneous fill and plain fill are drawn. The research shows that artificial fill of Xi'an is widely distributed, its thickness is mostly between 3 to 10 meters, maximum thickness in local areas can reach more than 10 meters. The occurrence and thickness of soil layers change rapidly in plane and their properties are complicated. Miscellaneous fill and plain fill are widely distributed in urban areas,depth of embedment is mostly within 3 meters, some areas can reach 3 to 10 meters, and very few areas can reach more than 10 meters.