The household characteristics in an electric grid include the socio-economic status of households, the dwelling properties, the information on the appliance stock, and so forth. These characteristics are significantly beneficial to electric retailers, because they can be utilized to provide personalized services, improve the demand response, and make better energy efficiency programs. However, these privacy-sensitive characteristics (e.g., employment, income, age of residents) require time-consuming surveys. Also, it is difficult to gather such residential household information in a large scale. In recent years, the increasing availability of electricity consumption data makes it possible to infer household characteristics from residential electricity consumption data. A number of supervised learning methods have been proposed. Among these solutions, features are extracted from the electricity consumption patterns, and the selected features are used to train a classifier or regressor. However, the existed methods depend on a single contributing model, which can be possibly undertrained. To achieve the optimal performance of classifiers for characteristics identification, we propose an ensemble framework based on bagging algorithms. With the proposed ensemble framework, the performance of characteristic identification has been improved.