Artificial Neural Network‑Based Prediction of Nipa Sugar Production in Sarawak, Malaysia

Abstract

Nipa palm (Nypa fruticans Wurmb) plays a vital role in the socioeconomic development of coastal communities in Sarawak, Malaysia, where its sap is traditionally used for sugar production. However, nipa sugar production is highly variable due to fluctuating environmental conditions, making reliable forecasting challenging. Current prediction methods rely on empirical observations and historical trends, which fail to capture the complex, nonlinear interactions between climatic factors and sugar yield. This study addresses the need for a more accurate forecasting model by developing a feedforward artificial neural network (FFANN) to predict nipa sugar production using daily data collected for twelve months from Kampung Tambirat, Sarawak. The FFANN model incorporates key environmental variables, including temperature, humidity, wind speed, atmospheric pressure and sap yield, and is trained using the resilient backpropagation (RPROP+) algorithm. The model’s performance was compared to classical time series models (ARIMA and seasonal naïve) and a decision tree regression model. Key results show that the FFANN model outperformed the benchmark models with a coefficient of determination (R2 ) of 0.73 and a normalized root mean square error (RMSE) of 0.097. Sensitivity analysis identified temperature, wind speed and humidity as the most influential factors on sugar production. The proposed FFANN model provides a robust decision support tool for nipa sugar producers, offering more accurate predictions for harvest planning and resource allocation. Future research could expand the robustness of the model by integrating multi-year datasets and IoT-based monitoring systems to enhance real-time forecasting capabilities.