Our data-driven prediction solution aims to extract information from mini-grids and households and optimize demand patterns by performing successive transformations of historical data to train powerful machine learning models that can cope with high uncertainty of demand patterns and subsequently minimize cost and energy consumption in newly encountered situations.