‘PRIDICTER’ - A Novel Comparative Analysis Approach for Price Predictions for Vegetable Markets in Sri Lanka integrating Explainable Techniques

Abstract

It is important to identify the price trends of vegetables to make better decisions in the retail and agriculture. However, vegetable prices fluctuate timely due to several factors, such as seasonality, perishability, demand and supply variations, customer preferences and also the availability and cost of raw materials which used in the cultivations of vegetables. Basically, the researchers use these factors into consideration when they work on identifying the patterns of fluctuation related to prices. The incomes of framers and retailers and also the satisfaction of customers are highly influenced by the regular changes of the prices of vegetables. Therefore, a proper price forecasting approach is needed for decision making purposes in the Sri Lankan vegetable market. In this study, the author has concluded price predictions using models such as SARIMA, XGBoost and LSTM with the daily retail price data for Beans collected from vegetable markets in Colombo from 2019 January to 2021 December. According to the evaluation RMSE, MAE and MAPE, LSTM outperforms compared to the other two models in predicting the retail prices for the Sri Lankan vegetable market. Apart from that, the user decided to add permutation feature importance as XAI technique to improve the interpretability of the black-box based model architecture which is used for the prediction purpose. Hence there are no studies in Sri Lankan retail sector focusing on applying explainable AI methods which helps human users to understand and trust the results and output created by the prediction model; this research can be known as a novel approach.