Comparison of Linear Regression and Polynomial Regression for Predicting Rice Prices in Lhokseumawe City

Abstract

Rice is a strategic food commodity in Indonesia, and its price fluctuations significantly impact inflation, economic stability, and poverty levels. Accurate price prediction is, therefore, essential for effective policymaking. The objective of this research is to develop a system for predicting the price of rice in Lhokseumawe City, employing a comparison of the accuracy of linear and polynomial regression models. To this end, daily price data from the Strategic Food Price Information Center (PIHPS) from 2020 to 2024 were utilized, with both models being implemented in Python. The findings indicate that 4th-order polynomial regression exhibited optimal performance, attaining a mean absolute percentage error (MAPE) of 1.85%, a mean absolute error (MAE) of 205.23, and a root mean squared error (RMSE) of 284.88. Conversely, the implementation of linear regression resulted in substantially elevated error metrics, with a mean absolute percentage error (MAPE) of 5.16%, a mean absolute error (MAE) of 553.91, and a root mean square error (RMSE) of 614.14. The findings indicate that 4th-order polynomial regression is a substantially more effective model for predicting rice prices in Lhokseumawe. The latter's superiority suggests that local rice price dynamics are characterized by significant non-linear patterns, rendering it a more robust tool for capturing data volatility and supporting data-driven policy.