Stockpredict makes a single-series forecasting experiment configurable through a command-line interface and notebook workflow. Users select a ticker, date range, lookback window and training settings. The pipeline downloads historical closing prices, constructs rolling windows and trains a stacked TensorFlow/Keras LSTM.
Chronological partitioning keeps later observations for evaluation, and the scaler is fitted only on the training segment. Predictions are converted back to price units for RMSE calculation and exported as dated CSV results and charts. The final observed window also produces one additional next-observation estimate.
Built with Python, NumPy, pandas, scikit-learn, yfinance and Matplotlib. The model follows a common learning pattern; the project focuses on making preprocessing, configuration and outputs inspectable. A naive forecasting comparator and retained comparative experiment records are proposed extensions. Predictive advantage and trading profitability have not been established.