HumanForYou — Employee Turnover Analysis

A machine learning project to identify and predict employee attrition at HumanForYou.

---

Table of Contents

• Background
• Objectives
• Project Structure
• Datasets
• Methodology
• Models & Evaluation
• Key Metrics
• Deliverables
• Installation
• Usage
• Ethics & Compliance
• Contributors
• License

---

Background

HumanForYou is a pharmaceutical company based in India employing approximately 4,000 people. The company is currently experiencing an annual employee turnover rate of 15%, which has a significant impact on:

• Productivity and institutional knowledge retention
• Recruitment and onboarding costs
• Team cohesion and project continuity

This project was initiated to understand the root causes of attrition and equip the HR department with data-driven tools to anticipate and reduce employee departures.

---

Objectives

1. Identify the key factors that contribute to employee attrition
2. Develop predictive classification models to flag at-risk employees
3. Interpret model decisions using explainability techniques (SHAP)
4. Recommend concrete, actionable strategies for HR to reduce turnover

---

Project Structure

HumanForYou/
├── data/                          # Raw data files
│   ├── general_data.csv           # Core HR and demographic data
│   ├── manager_survey_data.csv    # Manager performance assessments
│   ├── employee_survey_data.csv   # Employee satisfaction survey responses
│   ├── in_time.csv                # Daily arrival and departure times (2015)
│   └── out_time.csv               # Daily arrival and departure times (2015)
│
├── notebooks/                     # Jupyter notebooks
│   └── employee_turnover_analysis.ipynb
│
├── src/                           # Reusable Python modules
│   ├── data_loader.py             # Data ingestion utilities
│   ├── preprocessing.py           # Cleaning, encoding, feature engineering
│   ├── model_evaluation.py        # Evaluation metrics and comparison tools
│   └── run_analysis.py            # Runing the whole analysis

├── reports/                       # Project documentation
│   ├── ethics_document.md         # Ethical considerations and data governance
│   └── bibliography.md            # Academic and technical references
│
├── requirements.txt
└── README.md

---

Datasets

general_data.csv — Core HR Data

The primary dataset containing one row per employee with the following feature groups:

Category	Features
Demographics	Age, Gender, MaritalStatus
Job Information	JobRole, JobLevel, Department, BusinessTravel
Compensation	MonthlyIncome, PercentSalaryHike, StockOptionLevel
Experience	TotalWorkingYears, YearsAtCompany, YearsSinceLastPromotion, NumCompaniesWorked
Target Variable	Attrition (Yes / No)

---

manager_survey_data.csv — Manager Assessments

Manager-reported evaluations on a 1–4 scale:

Feature	Description
JobInvolvement	Employee engagement level as observed by the manager
PerformanceRating	Manager's rating of the employee's performance

---

employee_survey_data.csv — Employee Satisfaction Survey

Self-reported satisfaction scores on a 1–4 scale:

Feature	Description
EnvironmentSatisfaction	Satisfaction with the physical and social work environment
JobSatisfaction	Satisfaction with the role and daily tasks
WorkLifeBalance	Perceived balance between work and personal life

Note: This dataset contains NA values representing missing or unanswered survey responses. These are handled during the preprocessing phase.

---

in_time.csv & out_time.csv — Working Hours Data

Contains daily check-in and check-out timestamps for each employee throughout 2015. Used for feature engineering (e.g., average daily hours worked, overtime patterns, absenteeism indicators).

---

Methodology

The analysis follows a structured data science pipeline:

1. Data Loading & Exploration
        ↓
2. Data Preprocessing
   - Missing value imputation
   - Categorical encoding
   - Outlier detection
        ↓
3. Exploratory Data Analysis (EDA)
   - Attrition distribution
   - Correlation analysis
   - Department and role-level breakdowns
        ↓
4. Feature Engineering
   - Derived features from in/out timestamps
   - Overtime flags, average hours, absenteeism rate
        ↓
5. Model Development
   - Multiple classification algorithms trained and compared
        ↓
6. Model Evaluation
   - Cross-validation, confusion matrices, ROC curves
        ↓
7. Model Interpretation
   - SHAP values for global and local explainability
        ↓
8. Recommendations
   - Actionable HR insights based on findings

---

Models & Evaluation

The following classification models were trained and benchmarked:

Model	Description
Logistic Regression	Baseline linear classifier
Random Forest	Ensemble of decision trees, robust to overfitting
Gradient Boosting	Boosted ensemble for high predictive performance
Support Vector Machine	Effective in high-dimensional spaces
XGBoost	Optimized gradient boosting for tabular data

All models were evaluated using stratified cross-validation to account for class imbalance (the dataset is naturally imbalanced since attrition events are the minority class).

---

Key Metrics

Metric	Purpose
Accuracy	Overall proportion of correct predictions
Precision	Of predicted attritions, how many are correct
Recall	Of actual attritions, how many were caught
F1-Score	Harmonic mean of precision and recall
ROC-AUC	Model's ability to discriminate between classes
Feature Importance	Which features drive the model's decisions

---

Deliverables

#	Deliverable	Description
1	Jupyter Notebook	Full analysis pipeline with code, visualizations, and commentary
2	Ethics Document	Data governance, bias considerations, and responsible AI methodology
3	Bibliography	Academic papers and technical references used throughout the project
4	Presentation	20-minute presentation of findings and recommendations for HR stakeholders

---

Installation

1. Clone the repository

git clone https://github.com/<your-username>/humanforyou-turnover.git
cd humanforyou-turnover

2. Create and activate a virtual environment

python -m venv venv

# macOS / Linux
source venv/bin/activate

# Windows
venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

---

Usage

1. Place all data files in the data/ directory (see Datasets for expected filenames)

2. Launch the Jupyter notebook:

jupyter notebook notebooks/employee_turnover_analysis.ipynb

3. Run all cells in order — the notebook is self-contained and annotated at each step.

---

Ethics & Compliance

This project involves personal employee data and predictive modeling that could influence HR decisions. The following principles were applied throughout:

• Data minimization — only features relevant to the analysis were used
• Fairness — demographic variables (gender, age, marital status) were monitored for bias in model outputs
• Transparency — SHAP values are used to ensure model decisions are explainable and auditable
• No automated decision-making — model outputs are intended to support HR decisions, not replace human judgment

Full details are documented in reports/ethics_document.md.

---

Contributors

This project was developed by:

Name	Role
Manil Doudou	Developer
Maxime Moysset	Developer
Vanessa Cheptumo	Developer
Allexia Munene	Developer

---

License

Copyright © 2025 HumanForYou. All rights reserved.

This software is proprietary and confidential. Unauthorized copying, distribution, modification, or use of this software, in whole or in part, is strictly prohibited without prior written consent from HumanForYou.