Abstract:
Tea is the main agricultural value-added export product in Sri Lanka, providing
a significant contribution to the national economy. Specifically, green tea has
drawn worldwide attention in this sector due to its growing consumer demand.
Therefore, the main objective of this study was to obtain a reliable forecasting
model for green tea production using time series analysis, which is essential
for strategic planning to maintain competitiveness in the international market.
This study develops time series forecasting models for monthly green tea production
using both traditional Seasonal Auto Regressive Integrated Moving Average
(SARIMA) models and modern approaches such as Recurrent Neural Network
(RNN) models, specifically Long Short-Term Memory (LSTM) models. Monthly
production data from 2013 to 2024 were obtained from the Sri Lanka Tea Board
for analysis. Model forecasts were evaluated against observed values using standard
forecasting accuracy measures such as MAE, MAPE, and RMSE to identify
the best-performing model. SARIMA (2,1,2) (1,1,1)12 and LSTM models were
developed to forecast monthly green tea production for the test data months.
The dataset was split into training and testing data, focusing on capturing shortterm
production patterns. The results indicate that the LSTM model achieves
the lowest prediction error, demonstrating superior forecasting performance.The
results demonstrate that modern forecasting approaches, such as LSTM based
on Recurrent Neural Networks (RNNs), can effectively capture the nonlinear dynamics
of tea production and outperform traditional SARIMA models. These
findings suggest that industry stakeholders and policymakers can use RNN-based
forecasts to improve production planning, resource allocation, and market supply
management, thereby supporting more informed decision-making in the green tea
sector.
