234 points by codingfinance 6 months ago flag hide 16 comments
user1 6 months ago next
Interesting project, I've been looking into stock market prediction as well. How accurate is your model?
original_poster 6 months ago next
My model has an accuracy of around 70% on the test set. However, past performance doesn't guarantee future results.
user8 6 months ago next
70% accuracy is not a bad starting point, how did you handle overfitting?
original_poster 6 months ago next
I used early stopping, dropout, and batch normalization to prevent overfitting. I also used regularization techniques such as L1 and L2 regularization.
user7 6 months ago prev next
70% accuracy on a stock prediction model is amazing, I am skeptical about the results though. Can you provide a link to your code and dataset to reproduce the results?
original_poster 6 months ago next
Sure, I will publish the code and dataset on Github in a few days. Stay tuned!
user2 6 months ago prev next
70% accuracy is quite impressive. Can you share some details about your model and the data you used?
original_poster 6 months ago next
Sure, I used a 3-layer LSTM with 128 units in each layer, and I trained it on historical price and volume data. I also included some technical indicators as features.
user3 6 months ago next
Thanks for sharing the details. Did you consider using any other types of neural networks, such as convolutional neural networks (CNNs)?
original_poster 6 months ago next
Yes, I tried using CNNs, but they didn't perform as well as the LSTM model. I think the sequential nature of time series data is more suitable for LSTMs.
user4 6 months ago prev next
Have you thought about incorporating news data or social media data into your model? They might have an impact on stock prices.
original_poster 6 months ago next
That's a good point, I will definitely consider it in the future. I wanted to keep the model simple for now, but I plan to explore more complex models in the future.
user5 6 months ago prev next
How do you handle the large number of stocks and the huge dimensionality of the data? Do you train a separate model for each stock?
original_poster 6 months ago next
Yes, I train a separate model for each stock. I also use a dimensionality reduction technique, such as PCA, to reduce the number of features.
user6 6 months ago prev next
Have you considered using a transfer learning approach, where you pre-train the model on a large dataset of stocks and fine-tune it on a specific stock?
original_poster 6 months ago next
That's an interesting idea, I haven't tried that yet. I will definitely look into it. Thanks for the suggestion!