This repository contains the presentation, Jupyter workbooks and Flask/D3/Javascript application I created for the fifth project in Metis' Data Science program. It is a web application that, given location, size, and features, predicts the rent for a New York apartment.
I created this application to solve the problem where you are moving in New York and need to know what you should expect to pay for rent, how different apartment and building features affect prices, and what is a good value.
I built the application using a dataset of 95,000 listings I scraped from RentHop.com from March 25 - 27, 2017. From each listing, I extracted the location (longitude and latitude), size (number of bedrooms and bathrooms), price, and apartment and building features.
I stored the data in a MongoDB database running on and AWS instance, then used Python and the scikit-learn modules to test different linear regression models. The different algorithms were close in performance. For the application, I settled on an ElasticNet model with log10 price as the target variable.
The app folder contains the code for the Rent Report application that uses the MongoDB database to predict rents.
When you open Rent Report, it displays a heat map of all the listings in the city (giving you a quick sense of the relative prices in different areas), and a chart showing the distribution of all rents.
When you select an apartment size, it updates the chart to show the distribution of prices for just those size apartments.
When you enter an address or location, or click on an area of the map, it updates the chart to reflect prices in that vicinity, and predicts the rent for the specific size apartment in that specific area. From there you can toggle feature buttons to explore how different building and apartment features affect the prediction.