Fix CSV Prediction Errors in an Existing Python ML Script

I have an existing Python script that loads a trained customer-churn model and generates predictions from CSV files. The model works, but the script fails when optional numeric fields are empty and produces incorrect results when CSV columns are arranged differently from the training data. I need someone to correct the prediction-time preprocessing so the script follows the original training feature order and handles missing values consistently. The trained model must remain unchanged. I will provide the repository, trained model, saved preprocessing logic, original feature list, sample CSV, and expected class labels for ten records. The project already runs successfully in the supplied environment, and the issue is limited to CSV preprocessing and prediction. The delivery should include the corrected script, output CSV, two or three focused tests, and a short explanation of the changes. This does not include model retraining, accuracy improvement, API development, dashboard work, or deployment. Required: Python, pandas, and scikit-learn Timeline: One business day