E-Waste Quantification and Machine Learning Forecasting in a Data-Scarce Context

Quantifying e-waste in Sub-Saharan Africa remains constrained by scarce data, weak institutional reporting, and the dominance of informal sector activity. We present the first nationwide assessment of e-waste generation and Random Forest-based national forecasting in Sierra Leone. A mixed-methods survey administered 6000 questionnaires across all 16 districts, targeting households, institutions, enterprises, and informal actors. The study documented devices in use, storage, and disposal across the following six categories: ICT, appliances, lighting, batteries, medical, and other electronics. Population growth and device adoption simulations were combined with lifespan distributions and a Random Forest model trained on survey and simulated historical data to construct e-waste flows and forecast quantities through to 2050, including disposal fate probabilities for repurposing versus discarding. The results showed sharp spatial disparities, with Western Urban (Freetown) averaging about 10 kg per capita compared to 1.8 kg per capita in rural areas. Long-term district patterns were highly concentrated: 50-year annual averages indicated that Western Area Urban contributes 15.3% of national totals, followed by Bo (12.7%) and Western Area Rural (12.1%), with the top five districts contributing 59.1%. By 2050, total national e-waste entering reuse and disposal pathways was projected to reach 23.4 kilo tons per year (kt yr −1 ) with a 95% uncertainty interval (UI) of 11–42 kt yr −1 (and a 99% interval extending to 50 kt yr −1 ), corresponding to 0.9–3.4 kg/capita/year. Household appliances dominated total mass, ICT devices exhibited high reuse rates, and batteries showed minimal reuse despite high hazard potential. These findings provide critical evidence for e-waste policy, regulation, and infrastructure planning in data-scarce regions.