Distribution grids are evolving due to rising electricity demand and renewable energy integration, requiring efficient operation and effective planning. To achieve this, one essential step is translating the available load data into actionable insights. Machine learning (ML) approaches have emerged as promising solutions, leveraging increasing availability of data and computational capabilities. While research papers exist on applications of ML in power grids, a review in low-voltage substation-level is missing, an aspect that will be explored in this paper. The significance of emphasis at this level is twofold: ensuring privacy protection while gaining insights into consumption behavior, and eliminating the need for installing new meters or adjusting communication infrastructure. The paper covers three main ML algorithms, supervised, unsupervised, and reinforcement learning, their applications, while providing a critical discussion of their strengths and limitations. Furthermore, the paper provides recommendations for future studies.