Bitcoin Price Index Neural Network Prediction

Inspiration

The inspiration for this project came from high frequency trading platforms made by companies like Tradebot, the fact that lots of people have made money on the Bitcoin market in the past, and my desire to be a slightly less broke college student.

What it does

This project takes in historical Bitcoin price data and information from social media (currently just dummy data due the fact that most social media websites have paid api's for their historical data) and uses natural language processing to do sentiment analysis to generate the data set to create a model that can predict the way that the Bitcoin market will trend in the next day.

How I built it

This was built using 3 main things:

1. Python was the language used due to the speed of development it allowed for.
2. Tensorflow was used because it was easy to implement in a short time frame and provided a good model for the neural network.
3. Vader was used to do the sentiment analysis and natural language processing. It made the process for determining the sentiment values very easy to implement.

Challenges I ran into

Some of the biggest challenges I ran into were all related to my lack of experience with natural language processing and machine learning. Once I actually learned how to do machine learning the biggest challenge was to generate a model that was even slightly useful with data that was totally useless (thanks to the Twitter api for being really expensive).

Accomplishments that I'm proud of

The biggest thing I accomplished throughout this project was learning how to do machine learning in the span of around 12 hours. The next best thing would probably be that the model can actually predict the price trend of Bitcoin with a better accuracy than random guessing.

What I learned

I learned three main things:

1. Machine learning is not something to try to learn in a weekend.
2. Machine learning can be very very finicky.
3. Social media api's are expensive.

What's next for Bitcoin Price Index Neural Network Prediction

Following the success found in this project so far, I am going to implement automated paper trading to both feed data back into the model to improve it's accuracy, and to hopefully one day move into real trading with real money. This will (hopefully) be easier to train with the introduction of data from social media data sets (and not the dummy ones that I was forced to use now).

Built With

• keras
• python
• tensorflow
• vader