[68] - Monthly ForeCasht

• 

  Prediction for Apple

• 

  Prediction for Microsoft

• 

  Prediction for Tesla

Inspiration

Over the past 2 or so years, machine learning has overtaken most markets including image classification, generation, and segmentation; linear and non linear regression; and recurrent predictions based on time series data. Because of this, we decided to program an Long Short Term Memory model (LSTM) from scratch using TensorFlow to predict stock prices.

What it does

The LSTM uses long and short term memory to make daily predictions on stock prices. The way it does this is comparing the data from the previous day (short term memory) and data from many months prior (long term) and comparing the two to make a prediction.

How we built it

Using Tensor flow, we used the LSTM module from Keras and ran a training loop with 500 epochs that took in 100 data points per epoch. Next, we programmed an app with REACT to display the predictions comparatively to the existing data.

Challenges we ran into

The most difficult portions of the project were ensuring the quality of the data. By using validation during each epoch, we were able to verify confidence in the prediction since we used data that the model hadn't seen before to validate. Another challenge was designing the application to run smoothly and display the data correctly. Using the react js Line module, we were able to use arrays to produce the graphs seen on the app.

Accomplishments that we're proud of

Extremely proud to display data that is predicted with high confidence. In addition, we are excited to be able to format and display it in such a robust yet pleasing manner.

What we learned

What's next for [68] - Monthly ForeCasht

We plan on consistently updating the application so that it makes new predictions each week and learns based on how far off the predictions of the last week were.

Built With

• javascript
• python
• react
• tensorflow