HouseMates

• 

  App Overview

• 

  Predicted vs True

• 

  Residuals Histogram

• 

  Metrics

Inspiration

Seeing first hand the burden of students being taken advantage of by being charged unfair rent prices due to a lack of knowledge and experience of renting. This, coupled with the stress and logistics of finding housemates, led us to create this novel solution.

What it does

Our innovative system takes user preferences and helps match them with similar students, so they can form a house to their liking. Moreover, we utilised machine learning to regress fair rental prices so students can find good deals and be rest assured they aren't overpaying for rent- especially when every penny counts as a student.

How we built it

We used a React Native framework, along with JavaScript and Expo Go, to create a modern portable frontend. To facilitate the data processing we utilised Flask for API calls and Firebase to store and retrieve data. Python was also used to match students with similar preferences via Cosine similarity. Finally, we made use of Microsoft Azure to train various machine learning models, employing feature engineering for predicting fair rent prices. For the ML model, we used an ensemble of LightGBM, Extreme Random Trees and ElasticNet, all with StandardScalarWrappers after performing testing with numerous other models and hyper parameter tuning.

Challenges we ran into

Our group had limited experience with React Native, resulting in some parts of the app taking slightly longer than expected. Although, we were able to plan most things in advance, sometimes there was unintentional crossover in completing tasks. However, we dealt with this through regular checkin points throughout the 24 hours, to ensure we were on the same page with the same vision. We also underestimated the training time for the machine learning models and ran into an issue with poor accuracy due to a lack of feature engineering (one hot encoding) so we had to re-train our ensemble.

Accomplishments that we're proud of

This project allowed our groups members to each advance in various areas, from handling APIs, to Git management, and large-scale cloud applications of ML. We were particularly proud of balancing the bias-variance tradeoff of our Machine Learning model, as it performed better with the test set- indicating great generalisation. This was the majority of our groups first Hackathon, despite this it was a massive success and we all had a great time!

What we learned

We learned about the whole DevOps process, including how to implement Machine Learning in the cloud with remote training on a large scale concurrently with cluster computing; interacting with public APIs to get property data, for example ZooplaAI was used to collect student housing in Bristol and Bath. We also learnt resilience, in spotting hard to find bugs, and how to be patient when undertaking a project.

What's next for HouseMates

We would love to carry on improving the machine learning model, which currently has an accuracy of 86.8 and an R2 score of 0.877, to better the user experience of our app. We would also like to implement chat functionality, for students. In addition we would like to incorporate an LLM, to provide an AI real estate agent for the users.

Built With

• azure
• flask
• python
• react