Decentradata

Banner
Web Design

Inspiration

Decentradata was inspired by the rising popularity of Non Fungible Tokens. We were fascinated by the rising popularity of digital real estate within this virtual game called Decentraland. As aspiring data scientists, we decided to use our skills and create a prediction model around this enthralling virtual phenomenon.

What it does

Decentradata uses machine learning (Gradient Boosted Regression) to appraise digital real estate prices and find out the best investments that will provide the best returns 6 months from now.

How we built it

1) We calculated how many player-owned districts, plazas, and road parcels are present in a 100x100 area around a specific parcel. We used the Etherscan API to get all of the transactions ids and match them with Decentralands virtual land tokensId. The Ethetcan API returned all of the transactions of based on the requested information from the Decentralands API. All of the data was stored in MongoDB. 2) We dug into the previous land transactions (more than 1,000,000 data points) on ether’s blockchain, using etherscan.io, to find indications of whether a specific land/neighboring lands were over/undersold. 3) We used a density model to calculate the percentage of the infrastructure and visitor traffic to accurately model and predict the potential prices of the virtual properties. 4) We used a Gradient Boosted Regression to form an ensemble of decision trees to perform a regression model to predict and classify the potential prices of these lands in the future. 5) With this data, we built a model to find the best investments for a user and sent it over to the front-end built-on react.

Challenges we ran into

Creating a density model was one of the biggest challenges we ran into. We had to figure out the factors that can potentially affect virtual real estate values. We took inspiration from real-world factors to create this model for NFTs.

Accomplishments that we're proud of

We are extremely proud of the fact that we worked together as a team and computed 1,000,000 data points within 48 hours to create our model. Some of us worked with new databases and APIs to turn Decentradata into a reality while learning new skills at the same time.

What we learned

We learned that real estate prices, both actual and virtual are heavily influenced by their surroundings. The presence of nearby facilities like superstores, shopping centers, schools, hotels, etc has a big impact on the valuation of a property. We also learned to connect search trends to market demand and thus improving our model’s accuracy.

What's next for Decentradata

With Decentradata we plan on expanding our model to provide accurate predictions for many years to come. We hope that our model will help even newbies invest in virtual real estate and make sizable profits from their investments.