Thank you for the feedback! You raise a valid point about using highly correlated variables like past closing prices. In this case, the LSTM is likely capturing mostly linear relationships. To improve the model, we could incorporate less directly correlated features, such as price changes, technical indicators, or even external factors like volume or news sentiment. These could help the LSTM learn more complex patterns and non-linear relationships. This is definitely something worth exploring to enhance predictive accuracy and avoid over-reliance on simple past price data.