Skip to content

README.md

Latest commit

 

History

History
149 lines (105 loc) · 4.74 KB

README.md

File metadata and controls

149 lines (105 loc) · 4.74 KB

Fraud Analytics & Risk Scoring Dashboard

Project Overview

This project demonstrates an end-to-end fraud analytics and risk scoring workflow, from raw transactional data through rule evaluation, account-level risk scoring, and an operational summary dashboard.

The goal of the project is to replicate how fraud and risk teams:

  • Monitor suspicious activity using rules
  • Classify severity (HIGH / MED)
  • Aggregate risk at the account and customer level
  • Surface actionable insights for fraud operations and investigations

The final output is a Fraud Operations Summary dashboard designed for analysts, risk managers, and decision-makers.

Data Model

The project follows a star-schema-style layout with clear separation between dimensions, facts, and curated outputs.

Dimension Tables (data/raw)

  • dim_customers.csv – customer identifiers and attributes
  • dim_accounts.csv – account-level identifiers and metadata
  • dim_merchants.csv – merchant reference data
  • dim_devices.csv – device identifiers and attributes

Fact Tables (data/raw)

  • fact_transactions.csv – transaction-level data including international indicators
  • fact_login_sessions.csv – authentication and login session activity
  • fraud_rule_hits.csv – fraud rule triggers with severity and timestamps

Curated Outputs (data/curated)

  • account_risk_scores.csv – account-level risk scoring output derived from transactional and rule activity

Fraud Rules & Severity

Fraud rules are evaluated at the transaction level and recorded in fraud_rule_hits.

Each rule hit includes:

  • rule_code
  • rule_severity (HIGH / MED)
  • hit_ts (timestamp of rule trigger)
  • Associated account_id and customer_id

Severity is used throughout the analysis to differentiate:

  • High-confidence fraud signals
  • Medium-risk or investigatory signals

Risk Scoring Logic

Account-level risk is computed in the curated layer (Account_Risk_Scoring) using:

  • Frequency of fraud rule hits
  • Severity of triggered rules
  • Transaction behavior patterns (e.g., international activity rate)
  • Recency of suspicious activity

Key outputs include:

  • risk_score – composite numeric risk score
  • risk_tier – HIGH / MED / LOW classification
  • last_txn_ts – most recent transaction timestamp
  • last_hit_ts – most recent fraud rule trigger

Summary Dashboard

The Fraud Operations Summary dashboard provides a consolidated operational view of fraud activity.

Dashboard Sections

  1. Executive Fraud & Risk Overview
    High-level KPIs summarizing transaction volume, declines, rule hits, and severity distribution.

  2. Fraud Rule Activity Over Time
    Rule hit volume across recent monitoring windows (24 hours, 7 days, 30 days).

  3. Highest Risk Accounts (Model Output)
    Accounts ranked by composite risk score and risk tier.

  4. Fraud Rule Effectiveness & Severity Breakdown
    Rule-level performance showing total hits and severity mix.

  5. Account-Level Rule Activity
    Accounts generating the highest volume of fraud alerts, including recency and rule diversity.

  6. Customer-Level Rule Activity
    Aggregated fraud rule activity across customers for exposure analysis.

Screenshots

The following screenshots illustrate key dashboard views:

  • summary_kpis.png
  • top_risk_accounts.png
  • rule_effectiveness.png
  • account_rule_activity.png
  • customer_rule_activity.png
  • fraud_rule_activity_over_time_.png

SQL Structure

The SQL folder reflects a production-style analytics pipeline:

sql/ ├── 00_schema/ │ └── 001_create_tables.sql ├── 10_staging/ │ └── stg_transactions.sql ├── 20_rules/ │ └── rule_severity_logic.sql ├── 30_scoring/ │ └── account_risk_scoring.sql ├── 40_reporting/ │ └── fraud_summary.sql

This structure separates schema creation, staging logic, rule logic, scoring, and reporting.

Tools & Skills Demonstrated

  • Fraud analytics & rule-based detection
  • Risk scoring and tiering
  • Data modeling (fact / dimension design)
  • Google Sheets for analytical prototyping
  • SQL for schema, scoring, and reporting logic
  • Dashboard design for fraud operations
  • Data export and project packaging for analytics portfolios

How to Use This Project

  1. Review raw and curated datasets in the data/ directory
  2. Examine scoring and reporting logic in the sql/ directory
  3. Open the dashboard file in dashboards/ to explore fraud insights
  4. Reference screenshots for a quick visual overview

Notes

This project is designed as a portfolio demonstration of fraud analytics concepts and workflows.
All data is synthetic and intended for educational and professional showcase purposes only.