300 points by algo_trader 6 months ago flag hide 8 comments
fintech_fan 6 months ago next
Fascinating topic! I've been looking into the latest trends in using ML algorithms for predicting stock market trends. What do people think about LSTM (Long Short Term Memory) models particularly for time-series data analysis in this context? I'm curious to hear about other promising ML methods in this field.
quant_nerd 6 months ago next
Hey, I totally agree! LSTM models have been proven efficient for predicting stock trends due to their capability to learn from historical patterns. Though, I find that combining LSTM with GRU (Gated Recurrent Unit) models achieves better results. Looking forward to reading other comments and opinions on the topic!
stanford_y 6 months ago next
Hey fintech_fan, is it possible to bring your dataset here? I'd also like to test my own model based on your use case. Which platform and language are you using? I've been working with the TensorFlow library to create my algorithms.
fintech_fan 6 months ago next
I'm using Keras on top of TensorFlow for my ML experiments, and my data consists of historical OHLCV (Open, High, Low, Close, Volume) values of major stocks. Unfortunately, I cannot share it openly as it was provided to me with certain restrictions. I'm happy to see others interested in testing their solutions!
deep_learning_guru 6 months ago prev next
Hi! Both LSTM and GRU models can significantly improve prediction accuracy in stock market trends. However, I'd like to put attention towards CNN-QR (Convolutional Neural Network-Quantile Regression) models. They can process large datasets while requiring low computational cost and no complex preprocessing. Thoughts?
bitcoin_trader 6 months ago next
Can you explain in more detail how CNN-QR models can require low computational cost without complex preprocessing? I'm looking for efficient algorithms.
deep_learning_guru 6 months ago next
CNN-QR models are based on 2D convolutional layers. The dimensionality of the input (price) is only one, so the computational cost is much lower compared to LSTM-based models. In addition, the Quantile Regression layer directly predicts the quantiles or the distribution, which simplifies the processing.
machine_learning_veteran 6 months ago prev next
I've seen engineers combining ML models for stock prediction for a comprehensive approach. For instance, using LSTM for major trends and a simple linear regression model for noise reduction.