akeDataAnalyst/banking-fraud-analytics

Interactive fraud detection and risk analytics dashboard simulating Awash Bank S.C. operations in Ethiopia. Features real-time fraud prediction, channel/branch risk analysis, and filtered transaction exploration using synthetic data, machine learning (Random Forest), and Streamlit. # Awash Bank Fraud Detection & Risk Analytics Platform - LiveDashboard Link: (awash-bank-fraud-analytics-…) **An interactive fraud monitoring and real-time risk prediction dashboard simulating operations at Awash Bank S.C., Ethiopia's leading private commercial bank.** This portfolio project demonstrates end-to-end data analytics, machine learning, and BI development skills through a production-ready fraud detection system. ## Project Overview The platform provides comprehensive fraud insights across Awash Bank's nationwide branch network and digital channels (AwashBirr mobile transfer, ATM, POS, agent banking, bill payment, etc.) using a realistic synthetic dataset modeled on actual Ethiopian banking patterns. ### Key Achievements - **Synthetic Dataset Generation**: Created ~10,000 customers and ~100,000 transactions with authentic Ethiopian details: - Names, phone numbers, and real branch locations (Bole, Dire Dawa, Gondar, Mekelle, etc.) - Transaction channels, merchants, locations, and amounts in ETB - Engineered realistic fraud patterns (location mismatch, high-value transactions, off-hour activity) - **Exploratory Data Analysis (EDA)**: Identified key fraud indicators through univariate, bivariate, and multivariate analysis (fraud rate by channel, hour, amount, location mismatch). - **Machine Learning Model**: Trained a Random Forest classifier achieving high recall on fraudulent transactions using engineered features (amount, hour, channel encoding, location mismatch, high-amount flag). - **Interactive Streamlit Dashboard** (multi-page): - **Overview Dashboard**: Key metrics (total transactions, fraud cases, amount at risk) with visualizations for fraud rate by channel and top 15 risky branches - **Fraud Explorer**: Fully filterable transaction table (channel, branch) with "show only fraud cases" option - **Real-Time Fraud Predictor**: Input new transaction details for instant fraud probability and risk facto …