Early-warning model and prediction model for the side displacement of foundation pit were built in the aim of accurately grasping the deformation characteristics of foundation pit's side displacement in soft soil area. In the early-warning model, the early-warning discrimination indices were constructed using the multifractal detrended fluctuation analysis method, and then the change trends of the discrimination indices were determined by the Spearman rank test, hence the early-warning classification was completed. In the prediction model that is based on ridgelet neural network, the model parameters were optimized by rough set theory and trial and error method. Case study demonstrated that the early warning of the case in this paper was at level two, which indicated that the side displacement of the foundation pit tended to develop toward an unfavorable direction. Monitoring should be strengthened to improve the early warning for construction safety. In addition, the prediction accuracy and robustness of the ridgelet neural network can be effectively enhanced by parameter optimization, with the average relative error of the prediction results not exceeding 2%. The prediction results were consistent with the early warning results, which proved the accuracy of the analysis results.