|
|
| ---
|
| license: other
|
| library_name: sklearn
|
| tags:
|
| - binary-classification
|
| - tabular-classification
|
| - healthcare
|
| - physician-segmentation
|
| ---
|
|
|
| # Binary SEG_A vs SEG_B/C Classifier
|
|
|
| This repository contains the selected binary classifier for the first stage of a hierarchical physician segmentation strategy.
|
|
|
| ## Task
|
|
|
| Binary classification:
|
|
|
| - `0`: SEG_A
|
| - `1`: SEG_B/C
|
|
|
| The model predicts whether a physician belongs to `SEG_A` or should be routed to the second-stage `SEG_B` vs `SEG_C` classifier.
|
|
|
| ## Selected Model
|
|
|
| Best model: `HistGradientBoosting`
|
| Decision threshold for SEG_B/C: `0.45`
|
|
|
| ## Files
|
|
|
| - `best_binary_segA_vs_segBC.joblib`: trained model
|
| - `model_metadata.json`: model configuration and selected threshold
|
| - `binary_model_threshold_comparison_validation.csv`: validation threshold comparison
|
| - `test_predictions_binary_segA_vs_segBC_with_hcp_id.csv`: test-set predictions with HCP ID
|
|
|
| ## Notes
|
|
|
| The model uses flattened temporal tensors as input. Each physician is represented by weekly behavior across multiple features.
|
|
|
| The prediction probability `prob_SEG_BC` can be used to decide whether a physician should be classified as `SEG_A` or passed to the next B/C decision model.
|
|
|
