Geothermal heat flow is fundamental to understanding heat transfer and storage within the Earth's interior. However, it can only be measured at discrete locations, making continuous observation challenging and data acquisition costly. Consequently, the spatial distribution of available measurements is highly uneven. Traditionally, most heat flow maps have been constructed through direct interpolation of these measurement points, leading to significant biases due to uneven data distribution. Regions with dense measurements exhibit greater accuracy, whereas areas with sparse data suffer from reduced reliability in heat flow estimation. In contrast, machine learning techniques, when integrated with geological and geophysical parameters, provide a cost-effective and data-driven approach to generating accurate heat flow maps. This paper reviews the application of machine learning algorithms in geothermal heat flow prediction, systematically categorizing and summarizing existing methodologies. Key aspects of data preprocessing, feature selection, and model evaluation are examined, with particular emphasis on the impact of data quality and the selection of appropriate evaluation metrics. Additionally, the challenges associated with model underestimation are analyzed, and potential strategies for algorithmicimprovement and model optimization are discussed. In summary, advancing machine learning applications in heat flow prediction has significant implications for geothermal resource assessment, seismic hazard analysis, geodynamic research, and hydrocarbon exploration.