The heat increase caused by climate change has worsened the urban heat environment and damaged human health, which has led to heat-related mortality. One of the most important ways to respond to heat-related damage is to develop effective forecasting tools. However, accurately predicting heatwave damage is difficult in regions in a city with different conditions. Damage due to extreme heat can be evaluated differently in each region, as climatic, demographic and socioeconomic sectors are diversely distributed across local areas. In this study, we develop a random forest-based model for estimating the occurrence of heat-related mortality in a detailed spatial unit within a city. Through hyperparameter optimization, the model yielded accuracy, F1-score and AUC values of 90.3%, 94.75%, and 86%, respectively. The estimation results of the model were interpreted from the global and local perspectives by introducing the latest SHAP method. As a result of interpretation, demographic, socioeconomic and climatic sectors were determined to contribute the most to the estimation process. This is the first study of partial scenarios through the development and interpretation of a spatial unit machine learning-based occurrence estimation model for heat-related mortality.