Random Forest
- Hospital Data -

Before Data Transformation Snapshot

This is the original hospital dataset containing attributes like:

• Hospital Type, Hospital Ownership, ZIP Code, and various performance metrics. Many fields include "Not Available" or missing values, requiring data cleaning before modeling.

After Data Transformation Snapshot

• Features used include:

  • Count of MORT Measures Better (mortality)

  • Count of READM Measures Better (readmission)

  • One-hot encoded hospital types

• Target: Hospital Ownership (label encoded as y_cls)

•  Interpretation:

  • Feature space is sparse and unbalanced.

  • Several classes in y_cls are underrepresented (e.g., label 7 appears frequently, others rarely).

Results

Best Parameters:

• class_weight='balanced'

• max_depth=None

• min_samples_split=5

• n_estimators=50

Classification Metrics:

• High precision/recall for Government - Hospital District or Authority (1.00)

• Very poor performance on most other classes (recall and precision = 0.00)

• Overall accuracy: ~17.6%

• Weighted F1-score: ~0.12

 Interpretation:

Model is highly skewed toward one dominant class.

Class imbalance is a major issue despite using class_weight='balanced'.