Employee Attrition Prediction — HR Analytics

Problem Statement

Employee attrition costs companies 6–9 months of an employee's salary in hiring and training costs. This project builds a machine learning pipeline to predict which employees are at risk of leaving, enabling HR teams to take proactive retention measures before attrition occurs.

Business question: Can we predict employee attrition with enough accuracy and recall to make early intervention worthwhile?

---

Dataset

• Source: IBM HR Analytics Employee Attrition Dataset (Kaggle)
• Link: https://www.kaggle.com/datasets/pavansubhasht/ibm-hr-analytics-attrition-dataset
• Size: 1,470 employees × 35 features
• Target variable: Attrition (Yes/No) — 16.1% attrition rate (imbalanced)

Key features include:

• Demographics: Age, Gender, MaritalStatus, Education
• Job factors: Department, JobRole, JobLevel, OverTime
• Satisfaction scores: JobSatisfaction, WorkLifeBalance, EnvironmentSatisfaction
• Compensation: MonthlyIncome, StockOptionLevel, PercentSalaryHike
• Tenure: YearsAtCompany, YearsWithCurrManager, TotalWorkingYears

---

Project Structure

hr-attrition-analysis/
│
├── data/
│   └── WA_Fn-UseC_-HR-Employee-Attrition.csv
│
├── EDA.ipynb                        # Exploratory data analysis
├── preprocessing.ipynb              # Cleaning and feature engineering
├── modelling.ipynb                  # Model training and evaluation
├── run_pipeline.py                  # Execute notebooks in sequence
│
├── outputs/
│   ├── figures/
│   │   ├── attrition_by_dept.png
│   │   ├── feature_importance.png
│   │   ├── confusion_matrix.png
│   │   └── roc_curve.png
│   └── model_comparison.csv
│
├── requirements.txt
└── README.md

---

Methodology

Raw Data → EDA → Preprocessing → Feature Engineering → Model Training
         → Evaluation → Feature Importance → Business Insights

1. Exploratory Data Analysis

• Analysed distributions of all 35 features
• Identified class imbalance (16.1% attrition rate)
• Found key patterns: overtime, job level, and distance from home are strongly correlated with attrition

2. Preprocessing

• Dropped constant columns: EmployeeCount, Over18, StandardHours
• Label encoded all categorical variables
• No missing values found in this dataset
• Applied SMOTE to handle class imbalance in training set

3. Models Trained

• Logistic Regression (baseline)
• Random Forest Classifier
• XGBoost Classifier (best performer)

---

Results

model	accuracy	precision	recall	f1_score	roc_auc
Logistic Regression	0.881	0.7727	0.3617	0.4928	0.8045
Random Forest	0.8367	0.4615	0.1277	0.2	0.803
XGBoost	0.8673	0.7857	0.234	0.3607	0.789

Random Forest was selected as the final model based on not much difference ROC-AUC and F1-score.

---

Key Findings

Top 5 factors driving attrition (by feature importance)

Rank	Feature	Insight
1	OverTime	Employees working overtime are 3x more likely to leave
2	MonthlyIncome	Lower income brackets show significantly higher attrition
3	Age	Younger employees (under 30) have highest attrition rates
4	YearsWithCurrManager	Attrition spikes in first 2 years under a new manager
5	WorkLifeBalance	Employees rating WLB as "Bad" leave at 2x the rate

Business Insights

1. Overtime is the single biggest risk factor. Employees on overtime have a 30.5% attrition rate vs 10.4% for those who don't — a 3x difference. Reducing mandatory overtime should be the first intervention.

2. Early tenure is the highest-risk window. 40% of attrition happens within the first 3 years. Onboarding programs and manager check-ins during this period would have the highest ROI.

3. The model identifies 6 out of 10 at-risk employees correctly. With 1,470 employees, this means HR can proactively engage ~95 employees who would otherwise leave — at a fraction of the replacement cost.

---

Visualisations

Attrition by department

Top 10 features driving attrition

Model ROC curve

Confusion matrix (XGBoost)

---

How to Run

1. Clone the repository

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

2. Install dependencies

pip install -r requirements.txt

3. Download the dataset

Download from Kaggle and place inside the data/ folder: https://www.kaggle.com/datasets/pavansubhasht/ibm-hr-analytics-attrition-dataset

4. Run notebooks in order

jupyter notebook
# Open notebooks in order: EDA → preprocessing → modelling

5. Run full pipeline (optional)

python run_pipeline.py

---

Requirements

pandas==2.0.3
numpy==1.24.3
matplotlib==3.7.2
seaborn==0.12.2
scikit-learn==1.3.0
xgboost==1.7.6
imbalanced-learn==0.11.0
jupyter==1.0.0

---

Limitations and Future Work

• Dataset is relatively small (1,470 rows) — results may not generalise to all industries
• No temporal data available — cannot model attrition trends over time
• Future work: deploy model as a FastAPI endpoint so HR dashboards can query it in real time
• Could add SHAP values for individual-level explainability

---

About

Sumant Jadiyappagoudar Bioengineering graduate | Data Science & Computational Biology LinkedIn | GitHub | Email

---

Part of my data science portfolio. Other projects: SQL + Dashboard | A/B Testing | Pharma Analytics | NLP Sentiment Analysis