Sri Lanka tea auction price forecast by using data mining techniques

Abstract

This report presents the results and analysis of the research carried out to predict the auction price of tea in Sri Lanka. Sri Lanka Tea Board statistics say the price of tea often vary significantly with time of the year. The key factors that influence the price of tea are weather, climate change, plucking season, production capacity, export tonnage, US dollar exchange rate and crude oil price, as a result of that stakeholders in this industry are affected seriously and it also badly affects to the GDP of the country. Further, Sri Lanka Tea Board has no proper mechanism to evaluate these changes and predict tea auction price. There were a few research carried out to analyse and predict the price of tea based on the environmental and economic factors, hence this study may be beneficial for those who are involved in this industry. The factors related to tea price variations in the auction are overcome from investigating the correlation between the price of tea and the above key factors. This study used weekly tea auction price in the past seven years and prices & quantities of the above factors during that period. These past data were used to evaluate and identify the strength of correlation between the key factors and its variation pattern to predict the price of tea. The classification methods in data mining process were used for the predictions and was based on the analysis of correlation of key factors over the time period. Then regression models were used to forecast auction price. This analysis used several regression algorithm methods, but Regression by Discretization algorithm was identified as the best method among them to build an accurate prediction model. Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), Relative, Absolute Error (RAE) and Root Relative Squared Error (RRSE) were used to test and evaluate the accuracy of the results. These methods are not hundred percent correct to check the price variations due to fundamental drives such as human errors, natural disasters etc. However, this prediction model offer eighty five percent accurate, which is reasonable to forecast tea auction price.