Cost estimation is economically critical before starting off a construction project. One of the essential assignments for materials' prices prediction is to control the cost of inventory. Even though the prediction system based on support vector machine (SVM) recently has been emerged as a favourable choice, the prediction accuracy of SVM is usually deteriorated with nonstationary price data. Thus the way to explore workable price prediction still remains a challenge to be resolved for materials' cost control. In this paper, an enhanced online least squares support vector machine (LS-SVM) is proposed to predict the trend of building materials prices. Our design is to incorporate with empirical mode decomposition (EMD) to deconstruct nonlinear and nonstationary data for the set of intrinsic mode functions (IMFs), which are represented in sinusoidlike waveforms. Superior prediction, therefore, can be attained by predicting IMFs with online LS-SVMs. According to our simulation results, proposed EMD designs notably improve prediction accuracy from online LS-SVM and are workable for the cost estimation of building materials.