Sampling Techniques on Imbalanced Dataset

Assignment Overview

This project focuses on understanding the impact of different sampling techniques on imbalanced datasets and how these techniques affect the performance of various machine learning models.

The dataset used is a highly imbalanced credit card dataset, where the goal is to balance the data, apply multiple sampling strategies, and evaluate model accuracy.

---

Objectives

• Handle class imbalance using different sampling techniques
• Create balanced samples from the dataset
• Apply multiple ML models on each sample
• Compare model performance across sampling methods
• Identify the best sampling technique for each model

---

Dataset

• Source: Credit Card Dataset
• Link:
  https://github.com/AnjulaMehto/Sampling_Assignment/blob/main/Creditcard_data.csv

---

Tasks Performed

1. Loaded and analyzed the imbalanced dataset

2. Converted the dataset into a balanced class dataset

3. Created five different samples

4. Applied five sampling techniques:

   • Sampling1
   • Sampling2
   • Sampling3
   • Sampling4
   • Sampling5

5. Trained five machine learning models:

   • M1
   • M2
   • M3
   • M4
   • M5

6. Evaluated accuracy for each (Sampling, Model) pair

7. Compared results to find the best sampling technique per model

---

Accuracy Results

Model	Sampling1	Sampling2	Sampling3	Sampling4	Sampling5
M1	50.10	52.24	63.18	69.23	70.12
M2	59.25	65.27	68.72	28.36	30.25
M3	90.45	72.41	32.17	42.58	41.85
M4	78.25	56.24	47.23	33.44	40.12
M5	81.25	12.85	57.36	32.25	52.74

---

Key Observations

• Different models respond differently to sampling techniques
• No single sampling method is best for all models
• Proper sampling significantly improves model accuracy on imbalanced data

---

Technologies Used

• Python
• NumPy
• Pandas
• Scikit-learn
• Imbalanced-learn

---

Repository Structure

├── Creditcard_data.csv ├── sampling.ipynb ├── README.md

---

Conclusion

This project demonstrates the importance of selecting appropriate sampling techniques when dealing with imbalanced datasets. The choice of sampling strategy can greatly influence the performance of machine learning models.

---

Author

Mukul Ghai