In order to improve the accuracy of deformation prediction during construction of deep foundation pits, this paper proposes a support vector machine with genetic algorithm optimized parameters (GASVM) and autoregressive moving average (ARMA) model based on wavelet transform decomposition and reconstruction. This paper uses GASVM model to make one-step prediction and multi-step rolling prediction for trend items after wavelet decomposition, utilizing ARMA model to predict random items accordingly, and to sum the predicted values to get the final prediction result. Finally, taking a deep foundation pit of a subway station as a case, the prediction and analysis of the deep horizontal displacement of the supporting piles at the three monitoring points are obtained, and the short-term prediction value of one-step prediction and the medium- and long-term prediction value of multi-step rolling prediction are obtained. The predicted value of the GASVM model is used for comparison. The results show that the combined model in this paper effectively reduces the predictive error, and has achieved satisfactory results in both short-term and medium- and long-term estimations.