Data analysis and accurate forecasts of electricity demand are crucial to help both suppliers and consumers understand their detailed electricity footprints and improve their awareness about their impacts to the ecosystem. Several studies of the subject have been conducted in recent years, but they are either comprehension-oriented without practical merits; or they are forecast-oriented and do not consider per-consumer cases. To address this gap, in this paper, we conduct data analysis and evaluate the forecasting of household electricity demand using three realistic datasets of geospatial and lifestyle diversity. We investigate the correlations between household electricity demand and different external factors, and perform cluster analysis on the datasets using an exhaustive set of parameter settings. To evaluate the accuracy of electricity demand forecasts in different datasets, we use the support vector regression method. The results demonstrate that the medium mean absolute percentage error (MAPE) can be reduced to 15.6% for household electricity demand forecasts when proper configurations are used.