Cohesive soil in Gehu formation is the foundation bearing layer most used in the Suzhou–Wuxi–Changzhou area, and it's also one of the most commonly used stratum in underground space development. We selected 126 geological boreholes to research this stratum using 446 groups of sample data (400 groups of training data and 46 groups of prediction data). We used a neural network algorithm to establish a relationship model between physical parameters and mechanical parameters. We then input 46 groups of prediction data into the model to verify its reliability and also used specific data to research the relationship between physical and mechanical parameters. The results show that: (1) The Pearson correlation coefficient between the prediction results and measured data is more than 0.8, better than the fitting results of multiple linear regression. The model is verified as reliable. (2) There is a certain correlation between physical and mechanical parameters, and the neural network algorithm can express correlation models between multiple parameters. The neural network algorithm allows the understanding of the high value in the utilization of engineering geological big data. The method is simple and reliable, and it can be applied in the research and analysis of data.