Financial Anomaly Detection & Audit Dashboard
View on GitHubAudit Analytics
Financial Anomaly Detection & Audit Dashboard
SQL audit queries and an Isolation Forest model flag unusual activity for review across 217,441 transactions; flagged activity is not presented as confirmed fraud.
SQL (MySQL)PythonScikit-learnExcel VBAPower BI
Key Highlights
- 4,345 transactions flagged for review, representing INR 176.21M in exposure.
- Duplicate payments and high-value outliers surfaced; report formatting automated with Excel VBA.
- Five-visual Power BI dashboard covers exposure, merchant risk ranking and geographic distribution.
Tech Stack
SQL (MySQL)PythonScikit-learnExcel VBAPower BI
Overview
A transaction review workflow combining SQL audit queries, anomaly detection, spreadsheet reporting and dashboard analysis. The Isolation Forest uses transaction amount as its model feature and its flagged items require review.
Methodology
- Run MySQL queries for duplicate payments and other audit conditions.
- Apply Isolation Forest to transaction amounts with a 2% contamination setting.
- Export flagged transactions, format the exception report with Excel VBA, and summarize findings in Power BI.
Results & Findings
- 217,441 transactions analyzed; 4,345 (about 2%) were flagged.
- Flagged exposure totals INR 176.21M.
- The repository notes that the model uses amount only and depends on the selected contamination assumption.
Getting Started
The repository expects Python, MySQL and Jupyter. Its source dataset is external and is not included in the repository.
pip install -r requirements.txt
jupyter notebookProject Structure
- Notebook: analysis workflow
- audit_queries.sql: MySQL audit queries
- dashboard_screenshot.png: Power BI dashboard
- requirements.txt: Python dependencies
Screenshots
