Fraud Detection in Financial Transactions
Developed a machine learning model using Random Forest to detect fraudulent activities in over 6 million financial transactions. Included feature engineering, handling class imbalance with SMOTE, and data visualization for pattern analysis.
Transaction Anomaly Detection Using Machine Learning
Created a robust fraud detection machine learning model using Random Forest and SMOTE to handle severe class imbalance. The model was trained on over 6 million transaction records with engineered features like transaction amounts, balance differences, and temporal variables.
Conducted data preprocessing, handled missing data, and engineered features based on transaction steps, balances, and flags. Applied SMOTE to oversample the minority fraud class and trained a Random Forest classifier to successfully detect fraudulent transactions.
Overcame highly imbalanced data with fraud cases representing less than 0.2% of total transactions, making accurate classification difficult without oversampling and careful evaluation.
Implemented SMOTE oversampling technique and a Random Forest classifier to improve fraud detection. Visualized fraud patterns using bar plots and box plots to refine feature selection and model performance, achieving strong results.
Key Features
- Data preprocessing
- Feature engineering
- class imbalance with SMOTE
- Random Forest classification
- Data visualization for EDA
- Model evaluation