Abstract: With the development of urban engineering construction, the issue of construction engineering accidents has become more and more prominent. The geotechnical parameter interval obtained by using the traditional methods cannot meet the needs of actual engineering. Based on the idea of unsupervised learning, the peaty soil with the worst engineering properties is considered, and 8 physical indexes are selected as the input set. The principal component analysis (PCA) algorithm is used to realize the dimensionality reduction of multi-sample and multi-parameter decoupling, and the correlation and sensitivity of each physical index is obtained. Combined with its correlation and sensitivity, the comprehensive evaluation value of physical indexes of peat soil with different buried depths is given. The k-means clustering is used to analyze the relationship among physical index, and comprehensive evaluation value and engineering characteristics of peaty soil provide a theoretical basis for the selection of geotechnical parameters. The supervised learning method-BP neural network algorithm is used to analyze the unsupervised results and verify the accuracy of the (PCA—k-means) algorithm model. The normal samples obtained by clustering analysis are optimized by a variety of truncation methods to obtain a reliable value range, and the value results are compared with the actual engineering values to verify the rationality of the model engineering parameters. The algorithm model is of good engineering application value. The research results can provide references for engineering investigation, design and construction parameter values, and also provide a new analysis method for geotechnical parameter value analyses.