This model is a Temporal Fusion Transformer (TFT) trained for next-day Nifty50 return forecasting using a hybrid feature set of price dynamics and news sentiment.
It learns from:
- Ticker-level market features (OHLCV-derived signals and technical indicators)
- Multi-channel news sentiment (
direct, sectoral, global)
- Time-aware sequence context across multiple encoder windows
Training and evaluation are organized per window (7, 10, 15, 30), with checkpoints and metrics tracked for reproducibility.
The uploaded checkpoint corresponds to the selected best-performing window from our experiments.
This model is intended for research and educational use in financial ML workflows and is not financial advice.
