Machine learning for predicting volatile stocks

AI & Machine learning
1

Hello there :

I intend to create machine learning models for super volatile stocks… like MSGM (Motorsport Games inc.) , this stock went up more than 100% from prior close…on 17th april 2025 … this movements are just not regular… not many stocks go up that far in a single day … so I wonder…

In order to build a machine learning algorithm for predicting prices for this kind of stocks… (let’s say I want to predict the next price close…) what do I need to consider ?? what would be good inputs for my model ?? and what models should be more usefull for this ( Random forest, neural networks, etc. ) ??

Thanks

1 Like
2

Hi Ghery,

You’re absolutely right — these aren’t normal moves, and predicting them is less about regular price patterns and more about detecting potential breakouts.

How I think about it:

1. Return Distribution
  • Analyze the distribution of daily returns.
  • Even 2–5% daily moves can be treated as breakouts.
  • Reframe the problem as a classification task:
    • Will tomorrow’s return be > 2%?
2. Data Imbalance
  • Breakouts are rare (e.g., only 3% of days).
  • Use SMOTE or class weights to handle imbalance.
3. Model Selection
  • XGBoost / CatBoost: good for daily data
  • Neural Networks: need lots of data
  • SVMs: okay but don’t scale well
4. Inputs to Consider
  • Technical indicators: RSI, moving averages, volume
  • Price features: % change, intraday range, gaps
  • Volatility: ATR, historical volatility
  • Sentiment scores
  • Sector-wide activity
5. Beyond Traditional Features

Think less about exact prices and more about explosive move detection.

Thanks,
Ajay