Crop Price Prediction Using Machine Learning Approaches: Reference to the Sri Lankan Vegetable Market

Abstract

Every sector of this digital world is experiencing a noticeable change due to the development of information and communication technology, and the agriculture sector can be introduced as one sector that has undergone a remarkable revolution. Sri Lanka is a developing country in a tropical region and agriculture is the backbone of the nation, and plays a major role in the country's economy. The price that farmers receive for their harvest is critical to farmer satisfaction as well as the future survival of agriculture and primarily it depends on several factors such as demand, seasonal trends, and price offers from multiple suppliers. In recent years, crop prices in Sri Lanka have fluctuated drastically due to unpredictable climate change, natural calamities and many other circumstances. As the farmers were unaware of these uncertainties, they suffered huge losses in their harvest and became disillusioned and most of them intended to give up farming. Therefore, crop price forecasting seems to be a crucial factor in considering the future of agricultural production. Since the properties of crop prices are highly non-linear and combined with significant noise, forecasting crop prices is not an easy problem. Recently, many researchers have proposed various approaches for crop price predicting, among which data mining can be identified as an emerging approach that plays an important role in decision-making related to agricultural product price forecasting. However, in the context of Sri Lanka, there is no evidence of extensive studies on the use of data mining approaches for predicting crop prices, particularly for vegetables. The main objective of this research is to fill the gaps in the literature by assessing the predictability of vegetable prices in the context of Sri Lanka using data mining techniques. The variation in crop price was analyzed based on four factors namely rainfall, temperature, fuel price and crop production and experiments were conducted on four systematically selected vegetables covering up-country and low-country. Analysis was performed using five widely-used machine learning algorithms on similar phenomena and performance was evaluated using common evaluation metrics such as mean absolute error and root-mean-square error. Experimental results revealed that tree-based models are superior among the classifiers considered in forecasting vegetable prices in Sri Lanka.