You can run the app by going to this link: https://dubai-housing.azurewebsites.net/
BTW, as the famous saying goes: garbage-in garbage-out. I assume that the user has a common sense (for e.g. not selecting a studio apartment with 10 bathrooms!) If you input non-sensible data, you'll get non-sensible predictions. Currently the app doesn't protect from such inputs, but I'm aiming to think about how I can do this in future versions.
The script to scrape Bayut website can be found in scrape_bayut.py. This script scrapes all properties listed in the given Emirate. There is supported functionality to pass in whether you want furnished, unfurnished or all properties. However, in my experience, the number of furnished + unfurnished properties are not equal to all properties. Therefore, I resort to scrape all. Some of the data fields that are scraped include:
bedroomsbathroomsarealocationproperty_typeproperty_keyworksamenitiesdescription
Data cleaning, EDA and Feature Engineering is done in Cleaning, EDA and Feature Engineering.ipynb. For more information, please checkout this notebook. I tried to include a step-by-step process with relevant comments so you can easily follow my thought process!
Model training was performed using sklearn library. The notebook Model.ipynb contains all the relevant code for Linear Regression, Lasso Regression, Random Forest and Neural Network models. Just like before, you should be able to follow my thought process by going through the notebook.
I used Flask to develop the App, and Bootstrap for HTML elements and styling. The app is hosted on Azure.
This tool is developed solely for educational purposes and should not be taken as professional advice for property pricing purposes. I strongly discourage you to make any purchase or price-setting decisions based on this tool.
Please create a GitHub Issue and I'll get back to you ASAP!