| --- |
| library_name: sklearn |
| pipeline_tag: tabular-classification |
| tags: |
| - predictive-maintenance |
| - random-forest |
| - scikit-learn |
| --- |
| |
| # Predictive Maintenance Random Forest |
|
|
| This model predicts the binary `engine_condition` target from engineered engine sensor features. |
|
|
| ## Training |
|
|
| - Algorithm: `RandomForestClassifier` |
| - Source dataset: `saranka85/predictive-maintenance-engineered-data` |
| - Selection metric: mean stratified 5-fold CV F1 |
| - Best CV F1: 0.7564 |
| - Best parameters: `{"class_weight": null, "max_depth": 16, "max_features": "sqrt", "min_samples_leaf": 3, "n_estimators": 400}` |
|
|
| ## Held-out test metrics |
|
|
| - test_accuracy: 0.6642 |
| - test_precision: 0.6938 |
| - test_recall: 0.8364 |
| - test_f1: 0.7585 |
| - test_roc_auc: 0.6890 |
|
|
| ## Experiment tracking evidence |
|
|
| The experiment was tracked with MLflow using parent run `9bc671f6b04f4fc783913cc5a0dcd898`. |
| The portable reviewer bundle contains: |
|
|
| - `mlflow_runs.csv`: parameters, metrics, tags, and status for the parent and tuning runs |
| - `random_forest_cv_results.csv`: complete GridSearchCV results |
| - `classification_report.csv`: per-class held-out test metrics |
| - `confusion_matrix.png`: held-out test confusion matrix |
| - `experiment_summary.json`: experiment configuration and best-result summary |
|
|
| Reviewers do not need the local MLflow database; all run parameters, metrics, tags, and statuses are provided in portable CSV and JSON files. |
|
|
| ## Usage |
|
|
| Load `model.joblib` with `joblib.load`. Input columns and experiment details are recorded in `model_metadata.json`. |
|
|
