AI Stock Price Predicition

This was a personal project and the purpose was for me to implement machine learning techniques I have been learning about. In grad school, my focus is in controls. Several of my colleagues do research on adaptive control and we always talk about using existing time series machine learning techniques on stock data so I tried it.

When using machine learning to predict data that changes over time, the current state of the art method is to use recurrent neural networks. In this project I use Python and packages such as Pandas, Tensorflow, and Keras to implement a recurrent neural network (LSTM) that estimates stock price.

I used a relatively small amount of data as this project was mostly to learn Tensorflow, but I split my data into 80% training data and 20% testing data. I input the last 24 hours of price data into the network and the network will output its estimate to what the price will be in 5 hour. These time frames can easily be changed and the network can be retrained. The results  of this project were OK, I plan on revisiting this in the future and combining it with the my Stock Notification System Project.

I did this project in a Jupyter notebook and have included the notebook below.