Banking Customer Risk & Transaction Analysis

Overview

This project analyzes banking transaction data to understand customer behavior, transaction patterns, and fraud risk. The analysis focuses on identifying high-value transactions, evaluating fraud distribution, brewing risk insights, and deriving insights that support data-driven decision-making using SQL, Python, and Power BI.

---

Dataset

The dataset used is the PaySim Synthetic Financial Dataset for Fraud Detection, which simulates real-world mobile money transactions.

• Total dataset size: ~6.3 million transactions
• Working sample: ~55,000 transactions
• Key fields:
  • step – time
  • type – transaction type (PAYMENT, TRANSFER, CASH_OUT, etc.)
  • amount – transaction value
  • nameOrig – customer ID
  • nameDest – recipient ID
  • oldbalanceOrg, newbalanceOrig – sender balance
  • isFraud – fraud indicator (0/1)

---

Objective

• Validate and prepare transaction data for analysis
• Analyze customer transaction behavior
• Identify high-value transactions and spending patterns
• Detect fraud-prone transaction types
• Build a dashboard to track key metrics

---

Data Preparation

The dataset was structured but required validation before analysis:

• Sampled 55,000+ records from the original dataset for efficient processing
• Checked for duplicate records using SQL window functions (no duplicates found)
• Identified and handled negligible missing values
• Performed transaction consistency checks using debit–credit logic
• Created derived features such as:
  • Customer transaction frequency
  • Total spend per customer
  • High-value transaction classification

---

Analysis Performed

1. Transaction Distribution

• PAYMENT transactions accounted for the highest volume (~23K transactions)
• CASH_OUT and CASH_IN had significantly higher average transaction values

---

2. Customer Behavior Analysis

• Identified that ~18% of customers contributed ~62% of total transaction value
• Highlighted concentration of transaction activity among high-value users

---

3. Fraud Analysis

• Fraud was entirely concentrated in TRANSFER and CASH_OUT transactions
• TRANSFER had the highest fraud rate (~0.94%), followed by CASH_OUT (~0.33%)
• No fraud observed in PAYMENT, CASH_IN, or DEBIT transactions

---

4. Transaction Value vs Fraud Risk

• Low-value transactions had more fraud cases due to higher volume
• However, high-value transactions showed higher fraud probability (~0.20%) compared to low-value (~0.17%)
• Indicates increased risk per transaction for larger amounts

---

5. Data Consistency Insight

• Approximately 58% of transactions showed balance inconsistencies when validated using debit–credit logic
• Indicates limitations in balance tracking within the dataset
• Analysis focused on reliable features such as transaction amount and type

---

Dashboard

A Power BI dashboard was created to visualize:

• Transaction type distribution
• Fraud distribution and fraud rate by type
• High-value vs low-value transaction analysis
• Customer behavior and segmentation

Dashboard Preview

---

Tools & Technologies

• SQL – Data querying and analysis
• Python (Pandas) – Data validation and preprocessing
• Power BI – Data visualization and dashboarding

---

Project Structure

Banking-Customer-Risk-Transaction-Analysis/
├── data/
│   ├── raw_data.csv
│   └── cleaned_data.csv
│
├── notebooks/
│   └── data_cleaning_analysis.ipynb
│
├── sql/
│   └── analysis_queries.sql
│
├── dashboard/
│   └── powerbi_dashboard.pbix
│
└── README.md

---

Key Takeaways

• Transaction value is highly concentrated among a small group of customers
• Fraud is not random and occurs only in specific transaction types
• TRANSFER transactions carry the highest fraud risk
• High-value transactions have higher fraud probability per transaction
• Data validation revealed inconsistencies in balance fields, requiring selective feature usage

---

Future Improvements

• Apply fraud detection models using machine learning
• Perform deeper time-based trend analysis
• Enhance dashboard with additional filters and drill-down features

---

Contact

Ajay Tiwari
📧 tiwariajay110125@gmail.com
🔗 LinkedIn: https://www.linkedin.com/in/ajay-tiwari-849725284/
🔗 Portfolio: https://ajay-tiwari94-portfolio.vercel.app/