Galiishaq/ghana-telecom-data-generator

Synthetic telecom customer data generator for Ghana, calibrated to real market data (NCA, MTN, Census). Creates realistic churn datasets with regional variation for ML and analytics. # Ghana Telecom Customer Data Generator **A synthetic dataset generator for realistic telecom customer analytics in the Ghana market.** Built because I needed quality data to practice subscriber profitability analysis and revenue modeling, but couldn't find datasets that actually reflected African telecom markets. So I created one using real market statistics from regulatory and industry sources. --- ## Why I Built This When you search for telecom datasets online, you'll find the same few options repeated everywhere - most from European or American markets that don't reflect how telecoms work in Ghana or Africa more broadly. The distributions are off, the pricing doesn't make sense, and critical features like prepaid dominance and regional variations are missing or oversimplified. I wanted to work on subscriber profitability analysis, churn prediction, and revenue forecasting, but with data that actually represents the Ghana market. After spending time looking for quality datasets without much luck, I decided to build a data generator that creates realistic customer records calibrated to authoritative sources. ### The Approach I collected ground truth metrics from three main sources: 1. **National Communications Authority (NCA) Q4 2024 Statistical Bulletin** - Official regulatory data on market structure, usage patterns, and operator statistics 2. **MTN Ghana 2024 Financial Report** - Financial metrics and KPIs from Ghana's largest operator 3. **Ghana Statistical Service 2021 Census** - Population distribution across all 16 regions and urbanization rates I also did additional research on regional factors - network quality, economic conditions, and competitive intensity across Ghana's regions. This data is in `data/regional_factors.json` with full source citations. Using these sources, I built a probabilistic data generator that creates customer records with realistic relationships between features. The result is a 100,000-customer dataset that actually be …