An Automated Model for Predicting Weekly Tea Auction Prices at Colombo Tea Auction for Specific Elevations and Grades

Sri Lanka's tea business has existed since the 1800s and maintains a high standard worldwide. However, rivalry is increasing as more competitors enter the market, and tea exporters strive to make their operations more cost-effective. The discovery led to the development of a weekly tea price auction prediction model for a selected tea grade, Broken Orange Pekoe Fannings (BOPF), across different elevations. The existing research suggested using Box and Jenkins models over Machine Learning (ML) models such as Artificial Neural Networks (ANN) and Convolution Neural Networks (CNN). The research also suggested using external factors to improve accuracy. Supply, Demand, and Rainfall were the primary factors. This analysis required data from multiple sources to be combined into one dataset for model building. The tea auction data was from another source, but the rest was in the public domain. For the fourth elevation, the research group employed the SARIMAX variation of Box and Jenkins models to analyze them separately. The results ranged from a Mean Absolute Percentage Error (MAPE) of 0.26% to 6.03%, a manageable and acceptable level of risk. For structured Comma Separated (CSV) file inputs, a new function was created to easily extract predictions from the generated model, improving model usability.