Data Analytics for Maize Price Prediction using Regression Algorithms

Abstract

This research aims for finding the regression model that can predict maize price sold by Thai farmers. Three regression models are explored which are multiple linear regression, Ridge regression and Lasso regression. These algorithms learn from dataset collected by office of agricultural economics from Jan 2002-May 2019. We propose two new features which are the rate of cassava price change from 1 month and the rate of maize price change from 1-4 months. We do statistical analysis to see the relationship between features. Performance of regression algorithms are measured in terms of R-squared, Root mean square error and Mean absolute error. The experimental results reveal that feature selection play an important role for multiple linear regression with the R-squared = 0.94. We found that multiple linear regression outperforms Ridge regression (R- Squared = 0.86) and Lasso regression (R- Squared = 0.86). The mean absolute error of multiple linear regression, Ridge regression and Lasso algorithm are 0.31, 0.42 and 0.43, respectively. The root mean square error of these three regression algorithms are 0.50, 0.58 and 0.58, respectively.