The global deployment of smart meters has resulted in massive recorded data. By analyzing the data, we can capture users' electricity consumption behaviour. Data obtained contributes to the scheme design for the demand-side management (DSM) and the balance between the supply and demand sides in the power grid. For extracting consumption behaviour from smart metering data originally for billing purposes, a user classification model leveraging the ReliefF algorithm and CART method is proposed in this paper based on different DSM stimuli and time of use tariffs. Initially, 14 features strongly correlated to the electricity consumption patterns are extracted and weighted by the ReliefF algorithm, and 4 features therein are newly proposed in this work. Then, a Decision Tree is built by the CART method attributing the weighted features to classify users in various DSM program categories. Finally, $k$-fold cross-validation is carried out. Towards demonstrating the user classification feature correlations to various DSM programs as classification categories, the feature weights for the real-world ISSDA dataset are demonstrated and discussed. Experimental results and analysis for correlations of extracted features to DSM schemes and cross-validation accuracy prove that the proposed method is able to distinguish users under various DSM programs.