Deep Learning-Driven Predictive Modelling for Optimizing Stingless Beekeeping Yields

Abstract

Environmental factors like temperature, solar irradiance, and rain may influence the health and productivity of stingless bees. This paper aims to investigate the best approaches applied in meliponiculture to predict beehive health and products based on environmental variables and bee activity data. The data on temperature, humidity, rain, beehive weight, and bee activity traffic utilized in this project were monitored in real-time and saved on the Google Spreadsheet platform. The dataset extracted from the 6th of January 2024 to the 5th of February 2024, at a 15-minute time interval comprising a total of 2577 data points was analyzed using various deep learning approaches for best RMSE performance. A single-layer LSTM model with 50 units produced the best RMSE performance of 0.039, representing that the beehive weight was accurately predicted. This predictive capability can help farmers determine the optimum harvesting time based on weight forecasts, ensuring maximum yield and quality. Additionally, by providing early warnings of unwanted conditions such as swarming or potential attacks, this method significantly enhances the ability of beekeepers to take proactive measures to protect their colonies, safeguarding both bee populations and the livelihoods of farmers.

Keywords

• Melinponiculture
• deep learning
• LSTM
• RNN
• Stingless Beekeeping Yields
• Yields estimation