How to accurately predict peak traffic is difficult for various forecasting models. In this paper, least squares support vector machines (LS-SVMs) are investigated to solve such a practical problem. It is the first time to apply the technique and analyze the forecast performance in the domain. For comparison purpose, other two non-parametric predictors are selected because of their effectiveness proved in past research. Having good generalization ability and guaranteeing global minima, LS-SVMs perform better than the others. Providing sufficient improvement in stability and robustness reveals that the approach is practically promising.