Can news sentiment add a useful signal to historical price data? The stock_analyzer repository is an experiment that trains models to predict NVIDIA’s price direction and volatility using price trends and news sentiment.
The project includes a training script, a test script, and an interactive prediction tool. It gathers historical data and news sentiment, then compares multiple algorithms in an ensemble-style workflow and selects the best-performing model for the task. The README documents five years of training data spanning 2021–2025 and uses public data sources such as yfinance and Google News.
Treat the output as an experiment
A model can find patterns in a historical dataset without being able to predict the next market move reliably. That is especially important for a volatile stock like NVIDIA. This repository is for exploring features, training, and testing; its output is not financial advice, a guarantee, or a basis for trading decisions.
The project also separates direction prediction from volatility prediction, which are related but different problems. Testing both makes it easier to ask whether a model is learning a directional signal, a change in uncertainty, or merely quirks of its training period.
You can inspect the scripts, dataset structure, and plots in the NVIDIA stock analyzer repository. The useful lesson for me is that ML finance demos should make the data window and limitations just as visible as the prediction itself.
Top comments (0)