CSE Researchers Win MERCon 2026 Best Paper Award for Data-Driven Agricultural Volatility Forecasting
A team of researchers from the Department of Computer Science & Engineering, University of Moratuwa, has won the Best Paper Award in the Data Science and Artificial Intelligence track at MERCon 2026. The award-winning paper, "When Prices Double in a Week: Forecasting of Agricultural Volatility in Import-Isolated Markets," tackles a problem that directly affects food security in Sri Lanka: predicting sudden, severe price swings in a vegetable market with no import buffer to absorb supply shocks.
 |
 |
 |
The team, consisting of Ranuga Weerasekara, Heshan Nethmina, Manuja Ranathunga, Vinma Wettasinghe, Dinithi Navodya,Subavarshana Arumugam, Nirasha Munasinghe, Dr. Nisansa de Silva, Dr. Sandareka Wickramanayake, built a gradient-boosted ensemble model (XGBoost and LightGBM) trained on retail and farmer-gate price data spanning 2013 - 2019 across 12 vegetable varieties and 14 markets.
A key challenge was missing data. Urban hubs like Colombo have extensive retail records but no local cultivation, meaning zero farmer-gate prices. The team reverse-engineered the supply chain: for each such market, they identified the top two historical supplying producer regions for that vegetable and averaged their farmer-gate prices as a proxy, cutting farmer-gate missingness from 33.82% to 3.85%.
The model's feature engineering also traces each vegetable back to its actual growing region rather than the point of sale, using origin-averaged rainfall and temperature, since crops sold in one city are often grown hundreds of kilometres away. Diesel costs, exchange rate shifts, and the farmer-gate-to-retail spread were also engineered as predictors, with the spread proving the single strongest signal of future price movement.
The resulting model achieved 90.84% accuracy on held-out data and retained 85.96% accuracy on Sri Lanka's unseen 2024 hyperinflationary period without retraining, correctly tracking surges like a 274 LKR spike in Green Chillies, evidence the model learned real supply-chain structure, not just historical patterns.
The dataset and code are publicly available to support further research.
Paper - https://arxiv.org/abs/2606.29248
Dataset - https://huggingface.co/datasets/teamSynapse/sri-lanka-weekly-vegetable-retail-farmer-price-dataset
GitHub - https://github.com/RanugaVW/SL-Vegetable-Price-Forecaster
