Owing to the central role in electricity market operation, researchers have long sought to investigate the price responsiveness of both electricity supply and consumption sides. From the perspective of demand-side management (DSM), electricity prices prediction can be regarded as a pattern recognition problem of classifying future electricity prices with respect to a predefined threshold. From a fresh perspective this paper develops an efficient framework, called TSS-RFE-MRMR based SVM (Time series segmentation, recursive feature elimination, and minimum redundancy maximum relevance based support vector machine), for providing estimates of price fluctuation over certain valuation domains and modeling high-dimensional electricity market price without adopting additional impact factors. It starts from adopting a novel feature space determination scheme, called principal component analysis-dynamic programming (PCA-DP) based time series segmentation. Then, the RFE-MRMR filter for significant features selection is implemented, where both redundant and less relevant features are progressively eliminated among the potential feature sets. To test the performance of the proposed approach, it is evaluated on Ontario and New York electricity markets and compared with other method. Our experimental results indicate that the proposed approach outperforms other traditional method and present a relatively higher prediction accuracy on the electricity price.