The problem of access to clean water by urban poor households in low-income countries persists because the households are not connected to water distribution network (WDN). Hence, they lack public water supply in a poor sanitary environment, and are liable to suffer water-related diseases. This is especially true of the people of Nyanya-Mararaba Town in Nigeria’s Federal Capital Territory. This study explores the combination of geospatial technology and four machine learning techniques to develop functional models for predicting volume of water consumed by urban poor households. Random Forest performed better than other machine learning techniques during training with RMSE of 5.99 liter (L) in dry season, and RMSE of 6.59 liter (L) in wet season. Average water consumption in dry and wet seasons are 20 liters per capita per day (LCPD) and 23 LCPD respectively. The functional models were validated with RMSE of 5.67L and 4.92L in dry and wet seasons respectively, so providing a tool for planning water supply to urban poor households where there is no WDN.