Online Payments Fraud Detection with Machine Learning

Author: Leon Motaung

Environment: Python, Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn, Plotly

Project Overview

This project focuses on detecting fraudulent online payment transactions using machine learning. The dataset contains 284,808 transactions with 30 anonymized features (V1–V28), plus Amount, Time, and Class where 1 = fraud and 0 = normal. After cleaning, preprocessing, and feature engineering, the final dataset was saved as creditcard_final.csv.

Data Source

The original dataset is publicly available on Figshare: Credit Card Fraud Detection Dataset (external link).

Data Preprocessing

• Handled missing values
• Corrected data types
• Encoded categorical features (none present)
• Scaled Amount and Time features
• Removed outliers using IQR
• Added feature: scaled_amount (log-transformed)
• Saved cleaned dataset as creditcard_final.csv

Data Visualization

The following visualizations helped understand feature distributions and class imbalance:

• Boxplots: Detect outliers and feature spread.
  
• Histograms & Feature Distributions: Visualizing scaled features.
  
• Correlation Heatmap: Shows relationships between features and the target label.
  
• Scatterplot: Patterns between V2 and V4.
  

Baseline Model Evaluation

Note: Recall and F1-score are more important than Accuracy due to class imbalance.

Next Steps

• Apply SMOTE / undersampling for class imbalance
• Train advanced models: XGBoost, LightGBM
• Evaluate with Precision, Recall, F1-score, ROC-AUC
• Deploy via Flask or Streamlit for real-time fraud detection

Project Structure

• app.py – Training & prediction script
• draw.py – Visualization script
• creditcard_final.csv – Processed dataset
• Visuals – Boxplots, scatterplots, heatmaps, charts

This project gave me practical experience in preprocessing, visualization, and evaluating machine learning models for real-world fraud detection tasks.