Repository: Decision Tree HR Analytics Notebook analysed: Decision tree hr analystics 2 remodeled.ipynb

Develop a Decision Tree pipeline to predict employee attrition (left vs. stayed) using HR operational features, with a focus on:

• Robust preprocessing & hyperparameter tuning
• Balanced evaluation on imbalanced data
• Clear interpretability of results
• Business insights for HR retention strategies

• Source file (in notebook): HR_comma_sep.csv
• Shape (notebook output): 14,999 rows × 10 columns
• Features: satisfaction_level, last_evaluation, number_project, average_montly_hours, time_spend_company, Work_accident, left, promotion_last_5years, Department, salary

Attrition distribution (target = left):

0    11428  (76.2%)
1     3571  (23.8%)

⚠️ Class imbalance (~24% attrition) means metrics beyond accuracy are critical.

Key features such as satisfaction_level, time_spend_company, and last_evaluation show strong signals of separation between attrition and non-attrition groups.

Lower salary categories correlate with higher attrition risk.

satisfaction_level and time_spend_company emerge as the strongest predictors of attrition.

• Encoding:

  • Department → One-Hot Encoding
  • Salary → Ordinal (low < medium < high)

• Scaling: StandardScaler for numerical features

• Model: Decision Tree Classifier inside a Pipeline

• Tuning: GridSearchCV (StratifiedKFold, 5-fold) over depth, min_samples, and pruning (ccp_alpha)

• Grid search combinations tested: 768

• Best CV f1_macro: 0.9677

• Best parameters:

  • max_depth = 7
  • min_samples_split = 2
  • ccp_alpha ≈ 0.0039

Cross-validation results (mean ± std):

Accuracy:    0.9777 ± 0.0028
F1-macro:    0.9694 ± 0.0037
ROC-AUC:     0.9694 ± 0.0024

Metrics (test set):

Accuracy:   97.47%
Precision:  92.9%
Recall:     96.8%
F1-score:   94.8%

📌 Interpretation: The model catches nearly all employees likely to leave (high recall), while maintaining strong precision.

Top drivers of attrition from the tuned Decision Tree:

📌 Business Insight: Employee satisfaction and tenure dominate attrition risk — these should be the main levers for HR retention strategies.

The entire pipeline (preprocessing + model) is saved as:

hr_attrition_decision_tree_pipeline.joblib

Usage:

from joblib import load
pipeline = load("hr_attrition_decision_tree_pipeline.joblib")
preds = pipeline.predict(X_new)
probs = pipeline.predict_proba(X_new)[:, 1]

• Imbalanced target: handled with stratified CV, but requires ongoing monitoring in production.
• Model choice: Decision Trees are interpretable but can overfit. Ensembles (Random Forest, XGBoost) may provide further robustness.
• Feature scope: More signals (engagement scores, surveys, manager ratings) could improve accuracy and fairness.
• Ethics & privacy: HR data requires compliance with privacy regulations and bias audits before deployment.

# Clone repo
git clone <your-repo-url>
cd Decision-Tree-Hr-analystics

# Install dependencies
pip install -r requirements.txt
# or
pip install pandas numpy scikit-learn seaborn matplotlib joblib

# Run the notebook
jupyter notebook "Decision tree hr analystics 2 remodeled.ipynb"

📦 Decision-Tree-Hr-analystics
 ┣ 📜 Decision tree hr analystics 2 remodeled.ipynb
 ┣ 📜 HR_comma_sep.csv
 ┣ 📜 hr_attrition_decision_tree_pipeline.joblib
 ┣ 📂 images
 ┃ ┣ calibration plot.png
 ┃ ┣ confusion metrics.png
 ┃ ┣ Correlation metrics.png
 ┃ ┣ histogram plot of numerical columns.png
 ┃ ┣ Precision-recall curve.png
 ┃ ┣ ROC curve.png
 ┃ ┣ salary vs target.png
 ┃ ┗ top features that makes prediction.png
 ┗ 📜 README.md

• Developed and tuned a Decision Tree model for HR attrition prediction (15k records).
• Achieved 97.5% test accuracy with F1 ≈ 0.95 for the attrition class.
• Delivered interpretable insights — satisfaction and tenure are the strongest attrition drivers.
• Exported a ready-to-deploy pipeline (joblib), with reproducible EDA, evaluation plots, and business interpretations.

📌 This project showcases end-to-end ML workflow mastery — EDA, pipeline engineering, hyperparameter tuning, interpretability, visualization, and deployment readiness — highly relevant for Data Science & HR Analytics roles.