Text Features
- Pre-processed the extracted essay and deleted recurring tabs, spaces and newlines
- Word, sentence and paraghraph based statistical aggregations
- Punctuation statistics
- Punctuation mistake statistics (dot without a space after, comma with a space before, newline without a punctuation etc.)
Event Log Features
- Count and nunique features for events
- Statistical aggregations for all numerical columns
- Pause-time aggregations (for all text, pause between words, sentences and paragraphs, seperately.)
- P and R burst time and key-press count features
Training
- Used 10-Fold Stratified
- Used CatBoost, LGBM and XGBoost
- Hyperparameter-tuned all models with using Optuna
- Tracked OOF
- Ensembled final models using Optuna-OOF-tuned weights