Finomaly: An Open-Source Python Library for Financial Transaction Anomaly Detection

A modular Python library supporting rule engines and machine learning, helping users identify abnormal behaviors in financial transactions, suitable for fraud detection, anti-money laundering, and financial risk monitoring scenarios.

With the rapid development of fintech, anomaly transaction detection has become a core requirement for banks, payment institutions, and investment platforms. Whether it's identifying potential fraud, monitoring money laundering activities, or detecting system anomalies, efficient and accurate anomaly detection tools are key components of risk management systems.

Finomaly is an open-source Python library designed specifically for financial transaction anomaly detection scenarios. It adopts a modular architecture and supports both rule-based detection methods and machine learning models, providing developers with flexible and powerful anomaly detection capabilities.

The biggest feature of Finomaly lies in its dual-mode design, allowing users to choose the most suitable detection strategy based on specific scenarios:

Rule Engine Mode

The rule engine is suitable for handling known and clear anomaly patterns. Through a predefined set of rules, the system can quickly identify transaction behaviors that meet specific conditions:

• Large transaction threshold detection
• High-frequency transaction monitoring
• Abnormal time window identification
• Geographical location anomaly marking
• Account behavior pattern deviation detection

The advantages of this mode are strong interpretability, fast response speed, and no need for large amounts of historical data for training. For scenarios with strict regulatory compliance requirements, the rule engine can provide clear audit trails.

Machine Learning Mode

The machine learning mode is suitable for discovering complex and hidden anomaly patterns. By training on historical transaction data, the model can learn the distribution characteristics of normal behaviors and then identify transactions that deviate from the norm:

• Unsupervised anomaly detection (Isolation Forest, LOF, etc.)
• Supervised classification models (for known fraud-labeled data)
• Time-series anomaly detection (for transaction time series)
• Ensemble learning methods to improve detection accuracy

The machine learning mode can capture subtle anomalies that are difficult to define with rules, making it particularly suitable for dealing with evolving fraud methods.

Finomaly adopts a highly modular design concept, where each functional component can be used independently or flexibly combined:

• Data Preprocessing Module: Handles missing values, standardization, feature engineering
• Rule Engine Module: Defines, manages, and executes detection rules
• Model Training Module: Supports training processes for multiple ML algorithms
• Detection Execution Module: Unified anomaly detection interface
• Result Output Module: Generates reports, visualizations, and alert notifications

This design allows Finomaly to be used as an independent tool or easily integrated into existing risk control systems.

In scenarios such as e-commerce payments and credit card transactions, Finomaly can help identify:

• Fraudulent use: Multiple cross-location transactions in a short time
• Account takeover: Abnormal transactions after sudden changes in login behavior
• Fake transactions: Abnormal payment collection patterns on the merchant side
• Money laundering activities: Complex fund transfer paths

Financial institutions are facing increasing anti-money laundering regulatory pressure. Finomaly supports:

• Automatic identification of Currency Transaction Reports (CTR)
• Assisted generation of Suspicious Activity Reports (SAR)
• Dynamic adjustment of customer risk ratings
• Transaction network relationship analysis

In addition to security risk control, Finomaly can also be used for:

• System abnormal transaction identification
• Business process deviation monitoring
• Data quality anomaly detection
• Compliance audit assistance

Finomaly has a built-in complete data processing pipeline:

1. Data Loading: Supports multiple data sources such as CSV, JSON, and databases
2. Data Cleaning: Automatically handles missing values, outliers, and duplicate data
3. Feature Engineering: Extracts time features, statistical features, and aggregated features
4. Data Transformation: Preprocessing operations such as standardization, encoding, and dimensionality reduction