234 points by alexbrown 6 months ago flag hide 14 comments
finance_enthusiast 6 months ago next
Great job! I'm curious about the dataset and features used to train the algorithm. Did you use historical data or real-time data?
ml_developer 6 months ago next
Thanks! I used 10 years of historical data for training. Some features include open, high, low, close prices and volume.
another_user 6 months ago prev next
Any information on the validation method? How did you measure accuracy?
ml_developer 6 months ago next
I used k-fold cross-validation for the validation process. The model's accuracy was computed based on the predicted and actual prices with an accuracy of 95%.
critic_investor 6 months ago prev next
95% accuracy in a volatile and unpredictable market seems too good to be true, are there any restrictions or limitations on your approach?
ml_developer 6 months ago next
Yes, I need to clarify that 95% accuracy is based on in-sample testing only and there are indeed some restrictions and limitations to my approach. Real-world performance may have lower accuracy due to unforeseen events.
technical_analyst 6 months ago prev next
Can you share a bit more about the machine learning algorithm? Are you applying any specific technique for prediction like recurrent neural networks or support vector machines?
ml_developer 6 months ago next
I employed a Long Short-Term Memory (LSTM) based recurrent neural network housed within a deep learning framework to generate predictions.
ethical_investor 6 months ago prev next
How do you feel about the ethical implications of predicting stock prices? Could this be misused for insider trading or financial manipulation?
ml_developer 6 months ago next
I understand the concerns. My intention is to help make more informed decisions not to exploit the stock market. I strongly discourage any unethical practices.
trade_strategist 6 months ago prev next
What time horizon can this algorithm predict? Intraday, daily, weekly, or monthly?
ml_developer 6 months ago next
This algorithm is designed to predict daily and weekly price movements. However, I would advise using caution with these predictions.
data_explorer 6 months ago prev next
10 years of data might be too limited in scope. Have you experimented with data of a larger time frame like 30 or 50 years?
ml_developer 6 months ago next
A larger dataset would definitely add a wealth of historical knowledge, but I didn't have access to that much data for experimentation.