Soil compression parameters are significant factors for determining settlement to ensure safety of the civil engineering structures. These parameters are strictly evaluated through laboratory tests using samples collected from drilling boreholes. Currently, the testing procedures require both time and labour, and extensively increased construction’s cost in some cases such as in offshore structures. Therefore, it is necessary to establish a robust and reliable method, which can easily be used to obtain these parameters. In this paper, we employ Machine Learning (ML) models to relate field-testing results of piezocone penetration test (CPTu) to the compression parameters. A large database for this study is considered from five different construction projects, including two roads, a factory, and two massive container ports. Based on the database, a sensitivity analysis of the dataset and a feature analysis are performed. Furthermore, in order to select an appropriate ML method, several models are employed such as artificial neural network (ANN), deep neural network (DNN), DNN optimized with genetic algorithms (GA), and DNN optimized with particle swarm optimization (PSO). Comparisons between these ML models are also performed with remarkable better performances from the optimized DNN models than the classical models of ANN or DNN. Finally, validations are carried out using data from Nguyen Son (NS) road project to confirm the effectiveness and reveal the promising potential applications of the proposed methods.