Qamro/Prediction-Smartphone-Addiction-Submission-V2

Here's the enhanced version, honest summary of what actually moved the needle: What improved it: Feature engineering was the real driver: missingness indicators for every column (missingness itself carries signal here), plus ratio/interaction features like social_to_screen, sleep_deficit, weekday_weekend_diff, screen_per_age, etc. LightGBM with these new features: OOF AUC 0.9628 (up from 0.9620). Result: submission_v2.csv, same valid format (296,302 rows… See the full description on the dataset page: https://huggingface.co/datasets/Qamro/Prediction-Smartphone-Addiction-Submission-V2.