credit-card-fraud-using-SMOTE

Project Overview

• Here we, handle imbalanced dataset by applying SMOTE and remove the extreme outliers
• Compare which technique fits best to find feasible and resource efficient patterns in the dataset
• Logistic Regression, Decision tree, K-Nearest Neighbor, Naive Bayes, Support Vector Machine, Random Forest Regressors and XGBoost to reach the best model.

Code and Resources Used

Python Version: 3.9
Packages: pandas, matplotlib, seaborn, numpy, sklearn, xgboost, imblearn

Dataset

• Source : Kaggle - https://www.kaggle.com/mlg-ulb/creditcardfraud
• The datasets contains transactions made by credit cards in September 2013 by european cardholders that occurred in two days.
• It contains only numerical input variables which are the result of a PCA transformation. The only features which have not been transformed with PCA are 'Time' and 'Amount'.

Project Detail

Handling Imbalanced Dataset: For this project we use SMOTE (Synthetic Minority Over-sampling Technique) technique to handle imbalanced dataset.

Machine Learning Models: For this project I will implement the data to these models and determine which one fits the best:

• Logistic Regression
• Decision Tree
• Naive Bayes
• K-Nearest Neighbor
• Support Vector Machine
• Random Forest
• XGBoost

Model Parameter: To determine which machine learning model fits the dataset the best, I will use this parameter to decide:

• Precision-Recall
• F1-Score
• Accuracy

Step-by-Step Project:

Data Wrangling

• Scaling the 'Amount' and 'Time' columns
• Use SMOTE to handle imbalanced dataset and see the correlation of each columns
• Determine which feature have a negative and positive correlation
• Remove extereme outliers from features that have a high correlation with the classes

Data Analysis

• Decide which technique fits the best for the imbalanced dataset.

Handling Skewness

The data is heavily skewed therefore we use a power transformer to remove the skewness. Dataset before using power transformer- Dataset after using power transformer-

Model Building

• Use Logistic Regression, KNN, SVC, and Decision Tree as the model and compare each one of them

Results

The dataset is an imbalanced dataset with the class distribution like the above figure

Conclusion

1. The best technique to handle this imbalanced dataset is SMOTE
2. The best fitting model for this dataset is surprisingly KNN