There is a general consensus that the movement of electricity price is crucial for electricity market. As a practical tool to estimate the future prices, electricity price forecast is of great importance and use for the operations of market participants. So a hybrid forecast model is proposed in this paper that integrates independent component analysis (ICA) with least squares support vector machines (LS-SVM). First, a novel feature extraction method of price influence factors is proposed based on ICA, which aims at mining the latent source-features by using the higher-ordered statistical characteristics. After that, nonlinear regression modeling of electricity price and its extracted features is accomplished by LS-SVM with more efficient training and forecasting. Finally, Californian market data are employed to test the proposed approach.