youssefeladl/SMS-Fraud-Intelligence-Framework-Orange-Egypt-Internship-Project-

End-to-end fraud intelligence system for telecom SMS traffic. Combines quantile-based EDA, Isolation Forest anomaly discovery, clustering exploration, Random Forest classification, and a production-ready Streamlit dashboard for operator-level fraud detection. # 🚨 SMS Fraud Intelligence Framework **Orange Egypt – Big Data & AI Internship Project** qyhnfssnsdrcvb32huxrlx.stre… An end-to-end system for detecting and analyzing fraudulent SMS traffic at telecom scale. This project combines exploratory analysis, anomaly detection, supervised learning, and a production-ready dashboard — designed not just as a prototype, but as an **operator-level fraud intelligence tool**. --- ## 📌 Overview Billions of SMS messages flow through telecom networks daily. Hidden within them lies a fraction of traffic that represents fraud — small in volume, but significant in impact. This framework was built to uncover those invisible patterns using a **multi-layered approach**: - **Quantile-based Exploratory Data Analysis (EDA)** to identify anomaly thresholds. - **Isolation Forest** for unsupervised outlier discovery. - **Clustering** as a diagnostic instrument to validate anomaly structures. - **Random Forest** classifier trained on labeled anomalies, optimized for high recall. - **Streamlit dashboard** to operationalize detection with scalability and usability. --- ## 🔬 Methodology 1. **EDA as Foundation** - Performed quantile drilling (every 5%) across millions of SMS logs. - Determined 0.1% (≈0.001) as the logical anomaly ratio. 2. **Anomaly Discovery & Labeling** - Detected anomalies via Isolation Forest. - Used clustering to interpret anomaly distribution. - Converted signals into labeled fraud datasets. 3. **Modeling** - Trained a Random Forest classifier for supervised detection. - Prioritized **high recall** to minimize missed fraud cases. 4. **Deployment** - Built a Streamlit app with: - Multi-format support (CSV, Parquet, ZIP). - Sender-level aggregation & ranking. - Exportable anomaly reports. - Interactive charts for anomaly distributions. --- ## ⚡ Features - Scalable scoring for **multi-million row datasets**. - Hybrid fraud detection: **Isolation Forest + Random Forest**. - Operator-grade interface wi …