Abstract:
Purpose: Predicting the prices of crops is a principal task for producers, suppliers, governments and
international businesses. The purpose of the study is to forecast the prices of green chili, which is a cash
crop in Sri Lanka. Artificial neural networks were applied as they help to extract important insights from
the bulk of data with a scientific approach.
Research Method: The Time Delay Neural Network (TDNN), Feedforward Neural Network (FFNN) with
Levenberg-Marquardt (LM) algorithm and FFNN with Scaled Conjugate Gradient (SCG) algorithm were
employed on weekly average retail prices of green chili in Sri Lanka from the 1st week of January 2011 to
the 4th week of December 2018. The performance of models was evaluated through the Mean Squared Error
(MSE), Mean Absolute Error (MAE) and Normalized Mean Squared Error (NMSE).
Findings: Among the three methods implemented, the FFNN model using the LM algorithm exhibited the
highest accuracy with a minimum MSE of 0.0033, MAE of 0.0437 and NMSE of 0.2542. The model built
using the SCG algorithm fitted data with a minimum MSE of 0.0033, MAE of 0.0458 and NMSE of 0.2549.
Among the fitted TDNN models, the model with 8 input delays were a better model with an MSE of 0.0036,
MAE of 0.0470 and NMSE of 0.3221. FFNNs outperformed TDNN in forecasting green chili prices of Sri
Lanka.
Originality/ Value: The neural network approach in forecasting the prices of green chili provides more
accurate results to make decisions based on the trends and to identify future opportunities.
