Abstract:
The cinnamon industry in Sri Lanka, centered in the Galle District, one of the most important
economic pillars, represents over 85% of the world supply of Ceylon cinnamon. However, the industry faces challenges from unpredictable price changes due to climate variability and differences in grading standards. This study highlights these issues and their effects on farmers’ livelihoods and the competitiveness of the industry in the Galle District. A practical solution was developed, consisting of a machine learning model for price forecasting and an automated system for identifying cinnamon grades. For the price prediction model, we trained Support Vector Machine (SVM), Random Forest, XGBoost, and CatBoost using climatic and economic data, including rainfall, temperature, farm-gate prices, and exchange rates. We included seasonal trends as additional features. The results showed that fluctuations in exchange rates were the main factor affecting prices, while CatBoost provided the most accurate and stable predictions, with an R2 greater than 0.93. For grade identification, we created a computer vision pipeline that used color histogram and texture features from cinnamon quill images of the grades Alba, C5 Special, C5, and C4. A YOLO-assisted cropping process improved the model’s precision. We achieved high classification accuracy, with Random Forest reaching up to 87% accuracy, especially for cropped images. The integrated platform, now fully implemented as a mobile application, offers real-time price prediction and automated grade identification. This helps farmers and exporters make informed decisions, reducing economic instability and boosting Sri
Lanka’s share in the global market. However, the system’s reliance on historical data and its focus on the Galle District are current limitations. This solution shows how machine learning and computer vision techniques can improve sustainability and economic viability in the cinnamon industry.
