when developing solutions for social issues, it can be difficult to evaluate the impact they may have without a real world implementation. this may not be possible for reasons such as resource, time, and monetary constraints. to resolve these issues, simulation and machine learning models can be used to mimic reality and provide a picture of how these solutions would fare. in chapters 3 and 4, a deep learning approach to simulating homelessness populations in canada is presented. this model would provide policy makers with a tool to test different solutions for this societal problem without the need to wait for approvals or funding from local officials. in addition to this solution, data enhancement techniques are presented as a comprehensive dataset on homeless population transitions for such a model to learn from does not exist. lastly, chapter 5 presents a transfer learning architecture to detect tents in satellite images. the motivation for this work was that “tent camps” are common for homeless populations to live in and by having a solution to detect these from images, policy makers can easily see where to focus resources such as shelters for example. similar to the constraint present with the homelessness simulation, a comprehensive dataset on tents in satellite images does not exists. therefore, this chapter also presents a solution to generate an comprehensive dataset for the architecture to learn from. the result of this thesis is developed solutions to social issues that utilize the power of machine learning and simulation models.