Inspiration
-People at Atom Bank -Interests in data analysis, statistics, front/back end development, machine learning -Working to solve a real world problem -Programming black coursework -Free food -Sleep deprivation -Caffeine abuse -Friendships
What it does
-An automated valuation model for properties on a website -Takes user input data which includes post code, no. of rooms, type of house to estimate the value of a user's property -Uses machine learning to predict these values from data that we collected from available online source, which were Nestoria and HM land registry datasets -Multi-layer perceptron
How we built it
-Split it into 3 separate categories -The front end website where the user is able to enter in their data -The server which is a basic python flask server, it runs inference -The machine learning where user data was manipulated -Used web scraping on Nestoria to obtain additional data to the data that we got from the HM land registry data set -Using VS code and git collaboration w/ push & pull requests as required
Challenges we ran into
-Finding the data to run machine learning on -Legal issues with web scraping, rightmove & zoopla, both did not allow web scraping in their terms of service, however to obtain additional data ethically we were able to web scrape Nestoria who did not have such issues -However, while web scraping Nestoria, we had issues due to rate-limits so had to use mobile hotspots to bypass this problem -Integrating the 3 separate departments at the end, combining all the data -Trying to use a Python IDE on an iPad -Sleep -Caffeine Abuse
Accomplishments that we're proud of
-Having a solution that runs without errors -Learning new skills in machine learning and git collaboration -Not dying of caffeine overdose
What we learned
-Teamwork skills, being able to split the workload between multiple different team members -How to start machine learning in python -Creating a python server using the flask module
What's next for Proper-ly
-Creating a more robust overall model to predict the house prices (currently it it only 70% accurate) -Implementing additional factors into the machine learning approach such as distance to rail stations -We started on try to find the distance to between a post code and a rail station by using the pgeocode module in python (was not able to finish this unfortunately) -Trying to get additional data to use in the machine learning process eg. trying to obtain rightmove/zoopla api keys, which we were unable to obtain in these 24hrs
Log in or sign up for Devpost to join the conversation.