Location Location Location

Inspiration

We noticed that some factors in our data were heavily dependent on location, this allowed us to imagine an app that lets the user predict values such as the house price at a specific point using a neural network.

What it does

We take a single longitude and latitude coordinate, which we then use via a set of nested neural networks to generate a set of useful variables, with our final aim to produce a high accuracy price estimate.

How we built it

We used TensorFlow and keras to train a simple sequential model. We then created a set of functions that created each model based on the desired input variables and output variables. This was then used to plot over a map of the area showing the relative values our model predicted for a given position.

Challenges we ran into

We found that some variables were very useful for predicting our final desired variable of house price, such as "LSTAT". However, it was very hard to model for this variable, even with several other input variables included we were only able to achieve 80% accuracy for this value. As a result, we did not use this variable. In the end, this meat our house price predictions were accurate to within about 8%, rather than the 5% we managed when we included LSTAT data

Accomplishments that we're proud of

We managed to get a large amount of accuracy from very little input data. We were all new to this type of hackathon, and we've had a lot of fun learning along the way.

What we learned

We need to focus more on time management. We spent a lot of the first 12 hours of our hackathon without knowing our exact aims, and as a result, we didn't have time to produce a video presentation.

What's next for Location Location Location

We wish to expand the principles used in this app more widely. We plan to collect further data sets based on census data of cities in the UK, and attempt to find approximate prices of houses in areas in UK cities.

Built With

• keras
• python
• tensorflow