Goal: Predict minute-to-minute changes in the stock market and display relevant information in a browser interface.
- Alpaca API: Real-time prediction and analysis
- YFinance API: Model training and evaluation
- Socket.IO: Real-time system communication
The following was taken from my model training page where I fed my model data ranging from minute-to-minute data between 03/24/2025 - 03/29/25. This interval was specifically taken between this time to represent the week right before the testing week. Unfortunately, yFinance does not permit me to gather minute-to-minute data for over the past 8 days which is why I could only use this interval at most (I cannot go further behind).
The following was taken from my model deployment page in which I run a front-end application that displays a prototype of how the model would be applied. Ideally, the predictions would be fairly similar to the actual stock values however, I struggled to get the right scale for my model (-5 as opposed to -0.5) which is why, when my model predicts on the scaled value, the outcome is so skewed.
Throughout this project, I gained exposure to the more technical side of the stock market. I realized too late that my model design was not a good choice. Using an LSTM to predict stock prices, while a good idea, required me to create more modular code in that it must evolve as time passes. This would require different methods to save a model checkpoint and continue training it on new data so it can test on the unforseen.
- Using sliding-window technique in a real-world scenario Due to my model only being deployed in a prototype scenario, I never really got the chance to test it and train it during the actual market hours. What this means is that I missed multiple integrals functions that are required for this model to work in the first place:
- Code works in an infinite setting in that it will continuously grow without interruption
- Code only works in market hours
- Code works in the present