The research was based on the prediction of electricity consumption for household buildings from 16 U.S. states. The data collected was the energy consumed by 12083 households for a period of 6 months in the year 2009. The prediction was done for 3 class, 5 class and 10 class classifications. The CRISP DM standard was followed in the process of getting valuable insights from the data obtained. With the help of feature selection methods, we were able to select the important features for prediction and giving better results. An ensemble learner was proposed which had KNN, MLP Classifier and RF algorithms as base learners and XGBoost as meta learner. Classification accuracy and Confusion metrics were used as the evaluation metrics. The results were compared with the response obtained from the algorithms separately and with the result of Gradient boosting. Gradient boosting was found to give higher than other models.