Employee Absenteeism Predictor

Problem Statement

Employee absenteeism has a direct impact on productivity and efficiency. The goal of this project is to build a machine learning model that predicts whether an employee is likely to be absent for more than 3 hours on a given workday based on a variety of personal, professional, and health-related features.

The problem is framed as a binary classification task, where:

• 1 = Employee will be absent more than 3 hours
• 0 = Employee will be absent 3 hours or less

---

Workflow Overview

The project was developed across three stages, represented by three separate notebooks and a final module:

1. absenteeism_preprocessing.ipynb

• Initial data exploration and cleaning
• Feature engineering:
  • Extracted date-based features like Month and Weekday
  • Simplified Education levels
  • Grouped Reason for Absence into four broad categories (reason_type_1 to reason_type_4)
• Created the target variable Excessive Absenteeism (binary: absent > 3 hours = 1)

2. absenteeism_mL.ipynb

• Model training and evaluation
• Standardized select numerical features using a custom CustomScaler class
• Trained a logistic regression classifier
• Achieved an accuracy of 78% on the test set
• Saved the trained model and scaler using pickle

3. absenteeism_module_integration.ipynb

• Built a reusable Python module for deployment
  • CustomScaler: scales specified columns only
  • absenteeism_model: loads model and scaler, preprocesses new data, and generates predictions
• Loaded new test data and generated predictions:
  • Class probabilities
  • Binary predictions
  • Combined prediction results with preprocessed data

---

Features Used

Feature	Description
Reason for Absence	Categorical (grouped into 4 types)
Transportation Expense	Numeric
Distance to Work	Numeric
Age	Numeric
Daily Work Load Average	Numeric
Body Mass Index (BMI)	Numeric
Education	Binary (0 = high school, 1 = higher education)
Children	Count
Pets	Count
Month Values	From Date
Week of Day	From Date

---

Project Structure

├── absenteeism_preprocessing.ipynb
├── absenteeism_mL.ipynb
├── absenteeism_module_integration.ipynb
├── absenteeism_predictor.py
├── model (Pickled trained model)
├── scaler (Pickled custom scaler)
├── Absenteeism_data.csv (Training data)
├── Absenteeism_new_data.csv (Test data)
└── README.md

---

How to Use

1. Clone the repository and install dependencies (numpy, pandas, sklearn)
2. Run absenteeism_module_integration.ipynb to load new data and get predictions
3. The module can be easily plugged into a backend API or dashboard

---

Current Performance

• Test Accuracy: 78%
• Model: Logistic Regression
• Scaler: CustomScaler (only on selected numeric columns)

---

Suggestions for Improvement

1. Model Enhancements

   • Try more complex models: Random Forest, Gradient Boosting (XGBoost), SVM
   • Use grid search for hyperparameter tuning
   • Perform cross-validation for more robust metrics

2. Feature Engineering

   • Analyze interaction terms (e.g., Age × BMI)
   • Include additional features (e.g., total sick days, overtime hours, department)

3. Model Monitoring

   • Track model performance on new batches of test data
   • Add confidence thresholds for decision-making

4. Deployment

   • Convert module into a Flask or FastAPI app
   • Integrate with HR systems for real-time prediction

5. Explainability

   • Add SHAP or LIME to explain individual predictions

---

@kumathepanda