As the stock market has been a hot topic of conversation over the past few weeks, we wanted to try to apply techniques used in Machine Learning in order to gain insights into different economic trends that contribute to the price variations of a specific stock (ET) in the energy sector.
What it does:
Users are able to enter values for other stock prices, as well as an overall CCI value in order to predict the monthly change of ET's stock price.
How we built it:
We trained and tested multiple different regression models using various libraries such as sci-kit learn, pandas, and more. Data was gathered and cleaned through the employment of various written functions that would aggregate to an overall dataset. Then, preprocessing methods such as scaling and selecting the most influential feature variables through RFE further prepared our data for the final stages of training/testing using various regression models available through the importing of different libraries. Additionally, a recurrent neural network known as an LSTM model was created using Keras, in which the ET stock prices were aggregated and stored in memory cells in order to more accurately predict the monthly changes. Finally, through the employment of Flask which combined our Python code with HTML, we were able to create a small applet that allowed users to enter input data and receive an output ET stock price.
Challenges we ran into:
Aggregating data from various CSV files proved to be challenging, and creating a final dataset using different features proved to be challenging due to various formats and data sizes that needed to be accounted for in order to create a complete dataset. Additionally, in the LSTM model, the training/testing results were inconclusive due to the problem of an exploding gradient.
Accomplishments that we're proud of:
This hackathon was the first for three of our four group members, so our ability to create a working model through our hard work and diligence is something that we are all very proud of!
What we learned:
We learned a lot about collaboration as a team and utilizing each of our strengths in different aspects of the overall process in order to create our final product.
What's next for ET Stock Prediction:
Utilizing web-scraping techniques to aggregate more data to further improve accuracy, identify the cause of the exploding gradient problem for our LSTM, and taking advantage of other preprocessing methods such as GridSearch to identify key parameters and feature variables for testing/training.