Providing statistics about energy consumptions and accurate forecasts at appliances level can help energy clients and users to make informed decisions and reduce their electricity bills. This thesis presents a forecasting strategy that combines the benefits of load disaggregation, weather information, and ensemble learning to improve the accuracy of short-term load forecasts in a household. This thesis used combinatorial optimization to extract signature features of appliances from aggregated house energy consumption. The ensemble learning based on long short-term memory and the random forecast was designed to learn from the estimated energy consumption of appliances and predict the household energy consumption for three days ahead. The model was tested on a publicly available dataset, and the results showed that the load forecasting approach that considered the dynamic behaviors of home residents outperforms the accuracy of the benchmark forecasting methods that are trained with weather information or prior aggregated house power demand.