Heshan Nethmina

Ranuga Weerasekara, Heshan Nethmina, Manuja Mathiesha, Vinma Wettasinghe, Dinithi Navodya, Subavarshana Arumugam, Nirasha Munasinghe, Nisansa de Silva, Sandareka Wickramanayake

IEEE MERCon 2026

A gradient-boosted ensemble (XGBoost/LightGBM) with supply-chain-aware and seasonal features forecasts Sri Lankan vegetable price volatility, reaching 90.84% accuracy on held-out data and 85.96% on a fully unseen 2024 hyperinflationary period without retraining.

Heshan Nethmina

CS3111 — Introduction to Machine Learning Research Papers, University of Moratuwa

A controlled ablation within the SVM family for wine quality prediction finds that naive ordinal reformulation, feature selection, and class-weighting all underperform a well-tuned regression baseline, with only hyperparameter optimization giving a small but statistically significant improvement over the 2009 SVM-R benchmark.