Forecasting prices of salad vegetables using statistical and artificial intelligence techniques

This study aimedforecast the potential future prices of selected agricultural products using past price data. Multivariate time series forecasting plays a crucial role in various real-world domains, with price prediction being a particularly prominent research focus. However, this task is challenging as it requires capturing both intra-series and inter-series correlations. Existing literature mostly focuses on temporal domain correlations, neglecting other important factors. In this paper, we have collected a new dataset from the official website of the Turkish Ministry of Commerce. The dataset comprises daily prices and trade volume data for various vegetables, spanning a period of 1791 days from January 1, 2018, to November 26, 2022. Specifically, we have chosen parsley, lemon, cauliflower, and carrot as the target food items for price estimation in our study. After analyzing the acquired price data, we have employed widely used time series analysis models, namely Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and Random Forest (RF), to compare their performance. We evaluated these models on a 10-day price forecasting task for the four selected products. In general, all three models exhibited satisfactory results; however, the CNN model notably outperformed the others in terms of prediction accuracy.