Price prediction of pigeon pea (Arhar) using statistical and machine learning models in Rajnandgaon District of Chhattisgarh

Abstract

The present study aimed to forecast the monthly modal prices of pigeon pea (Arhar) in the Rajnandgaon district of Chhattisgarh using statistical and machine learning models. Twenty years of secondary data (2003-2023) were utilized from the Agmarknet portal, Directorate of Economics and Statistics, and meteorological sources. The study compared the predictive performance of the traditional Autoregressive Integrated Moving Average (ARIMA) model and the Support Vector Regression (SVR) model. Descriptive statistics revealed high volatility and seasonal variation in Arhar prices. The ARIMA model captured linear trends effectively but struggled with abrupt fluctuations. In contrast, the SVR model achieved superior forecasting accuracy with an R² value of 0.9460 compared to 0.4469 for ARIMA, and significantly lower MAE and RMSE values (42.4 and 35.67 respectively). Among SVR kernels, the Linear and Radial Basis Function (RBF) kernels performed best. These findings highlight SVR’s effectiveness in handling nonlinear price dynamics, offering a reliable forecasting framework to aid policymakers, traders, and farmers in informed decision-making and price risk management.