Below is an iframe containing an early Heroky Deployment of the project. As it is deployed in Heroku's free tier it may take a little while to load.
Below that is a description of the project, though you can probably get a good feel for it simply by exploring the Tableau visualizations in the Market Analysis tab, or playing with my Machine Learning algorithm in the Predictive Analysis tab.
Iframe navigation can get a bit clunky, so you can instead click the button below for a direct link to the deployed site.
This is an analysis of a nearly exhaustive data set containing residential real estate sales for St. Louis County from 2010-2020. The website has links to separate websites containing regional data for St. Louis but also contains Tableau visualizations of our data set on the Market Analysis tab. I also trained a Random Forest Regressor to predict home sale values based on a minimal number of parameters. The Random Forest Regressor is the smaller of two trained algorithms and the only one which could be deployed on Heroku. It only predicts within 5% of actual sale price about 25% of the time, making it not a great appraiser. The larger of the two models involved unsupervised clustering with a Gaussian Mixture algorithm and training different models for each cluster, resulting in a predictive pipeline which would first identify which cluster a property would be in and then use the respective model for that cluster. This has the benefit of a 50% increase in model accuracy as well as a dynamic estimation of its own confidence. This model was not deployed in the AI-AI project, however.
You can have the more lightweight Random Forest model predict for real or hypothetical homes in the St Louis County area on the Predictive Analysis tab. While is does make some pretty wacky guesses, the true value of this algorithm come from the change in price adding a bedroom, selling an identical house on a later year, etc. For example, my personal home was sold twice within our reference window. The second sale was several years after an addition which added square-footage, a bedroom, a bathroom, and several years to the year of sale. Appraisals for both sales were around 30% off, a pretty useless estimate, but the change in sale price was within 1% of the predicted change. So while you might not be able to trust the appraisal of your home, it may be worth it reference the change in value if you're considering adding a bedroom, bathroom, etc.