A machine learning project that predicts whether an employee is likely to leave the company. This system is trained on HR analytics data, carefully preprocessed, balanced using SMOTE, and deployed through a web interface for real-time predictions.
Employee attrition is a key challenge faced by organizations. By analyzing employee characteristics, job satisfaction, and work environment data, this project aims to identify employees who are at risk of leaving.
The system uses classification algorithms, primarily XGBoost and Random Forest, trained and tuned to handle class imbalance. It is deployed as a simple web application where HR professionals can upload employee data and receive predictions instantly.
- Source: IBM HR Analytics Employee Attrition & Performance Dataset
- Samples: 1470 employees
- Target Column:
Attrition(Yes/No) - Issue: High class imbalance (No: 1233, Yes: 237)
- Solution: Applied SMOTE (Synthetic Minority Oversampling Technique)
- Clean, reproducible ML pipeline
- Encoded & scaled input features
- Balanced dataset using SMOTE
- Hyperparameter tuning using GridSearchCV
- Final model: XGBoost, tuned and exported using
pickle - Simple web app (Flask) hosted on Render
- Accuracy: 92.43%
- Precision (Leave class): 93%
- Recall (Leave class): 91%
- F1-score: 0.92
Confusion Matrix:
[[345 24] ← Stayed
[ 32 339]] ← Left
The application is deployed on Render.
🔗 Live Web App: LINK
employee-attrition-predictor/
├── app/
│ └── app.py # Flask API to handle uploads and predictions
│
├── models/
│ ├── main_model/ # Contains Final tuned model
│ ├── scaler_models/ # Contains Saved StandardScaler
│ └── encoder_models/ # Contains Saved LabelEncoder
│
├── data/
│ ├── WA_Fn-UseC_-HR-Employee-Attrition.csv # Example input data
│ └── Test_Data_1_10.csv # Optional testing input
│
├── src/
│ ├── data_preprocessing.py # Encoding, scaling, input transformation
│ ├── model_loader.py # Load model from file
│ └── predictor.py # Make prediction from processed input
│
├── notebooks/
│ └── Test_Board.ipynb # Exploratory data analysis, Model training, SMOTE, tuning
│
├── requirements.txt # Project dependencies
└── README.md # This file
This project was built as a near-industry practice for our hackathon-ready ML team:
- 👩💻 Bidisha Pal – Data Pipeline, Preprocessing, Model Saving
- 👨💻 Ankur Halder – Exploratory Data Analysis, Testing, Input Validations
- 👨🔬 Rohit Gomes – Model Design, Tuning, Deployment, Project Lead
-
Clone the repository:
git clone https://github.com/YOUR_USERNAME/employee-attrition-predictor.git cd employee-attrition-predictor -
Install dependencies:
pip install -r requirements.txt
-
Launch the app:
python app/app.py
-
Open browser and go to
http://localhost:5000
- SHAP-based interpretability for predictions
- Excel input/output support
- Email alerts for high-risk predictions
- CI/CD pipeline with GitHub Actions
This project is developed purely for educational and demo purposes.
Thanks to the Kaggle dataset providers and the open-source ML community for tools and resources that made this project possible.
This repository also includes a fully interactive Power BI dashboard analyzing the same dataset used in the ML model.
📁 File: Employee_Attrition_Dashboard.pbix
- 📌 KPI Cards for Total Employees, Attrition Count, Retention Rate, and Average Tenure
- 🧩 Attrition breakdown by Department, Overtime Status, Education Field, and Job Satisfaction
- 📊 Clustered charts for Age vs Attrition and Overtime Impact
- 📅 Filter slicers for Gender and Marital Status
This dashboard offers a visual storytelling layer on top of the ML prediction engine and is helpful for HR managers and stakeholders.