#HR Attrition Analysis — Predicting & Preventing Employee Turnover

#Business Problem

"Why are employees leaving? Can we predict attrition early and target retention efforts?"

Data Science Approach

1. Convert to binary classification: Attrition = Yes/No
2. Three models for robustness & insight:
   • Logistic Regression: Interpretable odds ratios (HR-friendly)
   • Random Forest: Captures non-linear patterns (e.g., JobSatisfaction × Overtime)
   • XGBoost + SHAP: State-of-the-art prediction + explainable AI

Key Insights (From Results)

• Model performance:
  • Random Forest AUC: 0.811 → Strong predictive power
  • Logistic Regression: 0.805 → strong predictive power
  • XGBoost AUC: 0.779 → Slightly lower but more interpretable via SHAP
• Top 3 global drivers of attrition (SHAP):
  1. OverTime — Highest impact (mean |SHAP| = 0.62)
  2. MonthlyIncome — Second highest (0.44)
  3. StockOptionLevel — Third (0.38)
• Actionable insight: Employees working overtime with low income or stock options are at highest risk.

How to Run (100% Reproducible on Windows)

Prerequisites

• Windows 10/11
• Anaconda 2023.09 or later (includes Python 3.11.5)
  → Download Anaconda (64-bit)
  → During install: Check “Add to PATH” and “Register Anaconda”

Steps

1. Clone this repo
   Open Anaconda Prompt (search in Start menu) and run:

   git clone https://github.com/albertogoga/hr-attrition-analysis.git
   cd hr-attrition-analysis