A predictive–comparative framework for construction cost control using LSTM and digital twin technologies

Changes from v1.0.0 Dataset - Expanded synthetic dataset from 30 to 120 projects (838 → 3,183 monthly records)- Project IDs updated to P001–P120; test set: 24 held-out projects (stratified random split, seed 2025) Model files - Added cv_summary.json: aggregated results across all 25 fold–seed runs- Added model_config_example.json: representative training configuration (fold 0, seed 0)- Added test_projects.json: held-out test set IDs and split metadata Code - Updated core training and experiment scripts: bug fixes and reproducibility enhancements- Added Export_Descriptive_Stats.py: generates descriptive statistics table from synthetic dataset- Added Run_Learning_Curve.py: generates training and validation learning curves- Added generate_requirements.py: auto-scans source files to produce requirements.txt Dependencies - Added requirements.txt with exact pinned versions (Python 3.10, Windows 11) Documentation - Updated README.md and Dataset_Description.md to reflect expanded dataset and revised file structure