The deformation prediction accuracy of foundation pit is low nowadays. The denoising is carried out for deformation sequence of foundation pit by wavelet denoising and Calman filter, the trend term and error term are separated, and the trend term and error term is predicted by support vector machine (SVM) and BP neural network respectively. Meanwhile, the deformation trend of foundation pit is predicted by rescaled range (R/S) analysis so as to verify the feasibility of deformation prediction results. The case study shows that: 1) For wavelet denoising method, the denoising effect is superior; the average relative error and variance of the prediction results is 1.03% and 0.083 respectively; and the prediction accuracy is much higher. 2) The deformation sequence and deformation velocity sequence of foundation pit are prone to increasing, which coincide with the prediction results and verify the effectiveness of the prediction idea.