This analysis technique is all about extrapolating from incomplete data to make a more complete picture. It starts with the idea that we know something about how the data is collected and we have that collected data so we can use those two pieces of information to build a comprehensive view. Usually, this technique is applied to things like CT scans or MRIs; only recently has compressed sensing been used in the field of data science in a use case such as ours–we are on the leading edge. The key difficulty lies in the mathematical quantification of the CT/MRI sensing technology compared to the data collection methods in human surveys like those relating to homelessness.
Another key analysis method we use is multilevel methods. Multilevel methods are meant to handle non-independent samples and stitch them together into one more complete image of what is happening. There is promising work being done on police brutality and the murder of civilians that leads us to be hopeful this method can be used to produce better estimates on the homeless problem.
This method works hand in hand with each other method, building on patterns found by fixing the data with the above methods, and bringing in the data science below to make more abstract models. The main purpose of this modeling is to understand some concept or area well enough to quantify it mathematically and then interpolate and extrapolate new information. If we can predict what happens, we can use modeling to test policies and events. This would be very useful in confirming what we know to be true or to help try and fix things till we find something that works.
Disease Modeling
One suggested method of solving homelessness is to treat it like a disease, something that you can prescribe treatment to, and it seems to work. With that thought, we think it is possible to model homelessness as a disease using the already existing techniques to make a model that can help us understand how homelessness works on a population.
Stochastic Modeling
Stochastic Modeling is the technique of modeling with built-in unknowns. Things like weather and finances are some key examples of the power of stochastic modeling. In a world where we don’t know exactly what might happen, stochastic modeling tries to build that element of unknown into the model itself.
This is perhaps the widest net method. It includes things as simple as simply creating visualizations that make data more readable, to using machine learning and data mining to create a word mining AI to predict possible spikes in service demands using keyword searches. This aspect of the program is perhaps the most nebulous because it is also the most flexible and includes many small techniques whose veracity has yet to be determined in the context of our problem. That said, there are many well-established data science applications and we hope to bend as many of them to uses similar to the example above to help us understand what exactly homelessness is.
