anwoy71/Nifty50-prediction — reverse-engineered prompt

Build me a small finance ML project that predicts whether the NIFTY50 will close up or down the next trading day. I want it done in a careful, realistic way, not a flashy overfit backtest. Please use the market data for NIFTY50, Bank Nifty, and India VIX, clean and align it by date, create a target for next day direction, and engineer sensible time series features like short term returns, momentum, volatility, moving average relationships, VIX regime style signals, and cross market relative strength.

Then train a binary classifier with proper time aware validation. Please avoid data leakage, use chronological splits and walk forward testing, and do not shuffle anything. I also want a simple trading style evaluation with threshold analysis and metrics like out of sample accuracy, balanced accuracy, AUC, hit rate, Sharpe after costs, and max drawdown.

Keep the workflow reproducible and easy to follow, ideally in a few notebooks with processed datasets saved out. If needed, look up current docs online.