This project focuses on detecting anomalies in financial transactions using machine learning techniques. By analyzing transaction patterns, we can identify suspicious activities that may indicate fraud. The project applies Isolation Forest for anomaly detection, supported by data visualization.
-
Objective: Identify suspicious financial transactions that deviate significantly from normal behavior.
-
Techniques:
- Data exploration and processing
- Anomaly detection using Statisticl methods and Machine Learning model
- Model training and evaluation using Isolation Forest
- Balancing data using SMOTE to improve model robustness.
-
Tools & Libraries:
- sklearn, imblearn, Matplotlib, Seaborn, Plotly
- Anomaly Detection Rate: 2% of transactions were flagged as anomalies.
- Isolation Forest effectively detected outliers with 99% accuracy on the test set.
• Transaction Amount Distribution: Histogram of transaction amounts to understand data distribution.
• Boxplot Analysis: Insights into transaction variations by account type.
• Scatter Plots: Detecting patterns in anomalies based on transaction attributes.
• Correlation Heatmap: Relationship analysis between numerical features.