Employee Churn Prediction with Machine Learning

• This project uses machine learning to predict whether an employee will leave the company based on HR records.
• It includes complete data exploration, preprocessing, feature engineering, Building pipelines, Model selection & hyperparameter tuning, data leakage prevention , evaluation, and deployment using a Streamlit web app.

Demo

Or try the live app 👉 Streamlit Cloud Link

Understand the business scenario and problem

Project Goal:

• Help HR to identify employees at high risk of leaving the company.
• Provide actionable, data-driven insights to improve retention.
• Build a reliable, interpretable, and production-ready ML system using scikit-learn pipelines and Streamlit.

Project Highlights

• Best Model: Tuned Random Forest Classifier
• Performance: AUC = 93.8% | Accuracy = 96.2% | F1 Score = 88.67% | Recall = 90.36 %
• EDA: Identified key attrition drivers such as project overload, overwork, tenure, number of projects
• Feature Engineering: Introduced a binary overworked feature: employees working >175 hours/month.
• Data Leakage Prevention: Removed satisfaction_level and average_monthly_hours(as these may reflect post-decision bias in the company).
• ML Pipeline/Tuning: Used Pipeline, ColumnTransformer, and GridSearchCV
• Deployment: Interactive Streamlit app for real-time risk prediction and analysis

Folder Structure

─ data/                   # Processed dataset
─ models/                 # Serialized model pipeline
─ notebooks/              # Jupyter notebooks (EDA + modeling)
─ src/apps/app.py         # Streamlit app entry point
─ src/pipeline.py         # Pipeline + tuning logic
─ src/apps/predict.py     # Churn prediction form logic
─ src/apps/eda.py         # EDA visualizations
─ src/apps/utils.py       # Load data and model
─ README.md

Setup Instructions

1. Clone this repo:

git clone https://github.com/your-username/employee-churn-prediction.git
cd employee-churn-prediction

2. Install dependencies:

pip install -r requirements.txt

3. Run the Streamlit app:

streamlit run src/apps/app.py

---

Model Metrics for Best Selected Model(Random Forest)

Metric	Score
Accuracy	96.2%
Precision	87%
Recall	90.36%
F1 Score	88.67%
ROC-AUC	93.84%

Best Model: The Random Forest model demonstrated the strongest performance.

Key Metrics:

• AUC: 93.8% - This indicates an excellent ability to distinguish between employees who will leave and those who will stay.
• Precision: 87.0% - This shows when the model predicts an employee will leave, it's correct around 87% (or more) of the time.
• Recall: 90.36% - The model successfully identifies about 90% (or more) of employees who actually end up leaving and also shows strong identification of at-risk employees.
• F1-Score: 88.7% - This shows a strong balance between precision and recall.
• Accuracy: 96.2% .Overall, the model makes correct predictions (leave/stay) for a very high percentage of employees.

Summary of model results

Logistic Regression

The logistic regression model achieved precision of 80%, recall of 83%, f1-score of 80% (all weighted averages), and accuracy of 83%, on the test set.

Tree-based Machine Learning

After conducting feature engineering, the decision tree model achieved AUC of 93.8%, precision of 87.0%, recall of 90.4%, f1-score of 88.7%, and accuracy of 96.2%, on the test set. The random forest modestly outperformed the decision tree model.