The tracking of electricity infrastructure locations is crucial to making informed decisions on grid expansion and energy supply alternatives. However, in developing settings, these tasks are limited by technical and budget capacity constraints where the most recent data on the locations of low- and medium-voltage grids is outdated or even unknown. Currently, utilities in high-income economies monitor these lines using sophisticated sensing devices, airborne laser scanning, and field surveys which are unaffordable in emerging economies. In this work we aim to improve upon an existing open-source electricity mapping tool that uses night-time light data as the main proxy of electrification. Using ground-truth data from Kenya, we validate the performance of the existing tool and proposed a learning model to improve the detection of electrified sites. Our results show that our learning model is able to correctly identify ≈78% of those places which had electricity but were not identified before and improve the detection accuracy by up to ≈7%. Moreover, we show that using daily composites of nighttime data combined with other open-source data sources significantly helps the generation of accurate electricity access maps.