Machine Learning and ARIMA Models for Spatiotemporal Analysis and Forecasting of Shrimp Yields across the Aquatic Systems of Southern Coastal Bangladesh

Abstract Shrimp aquaculture is vital to Bangladesh’s economy but faces challenges from environmental degradation, climate variability, and resource management. This study analyzed secondary data (2001–2024) from four southern coastal districts-Khulna, Satkhira, Bagerhat, and Jashore-to assess spatiotemporal shrimp yield patterns using Multiple Linear Regression, Random Forest Regression, and ARIMA forecasting. Random Forest showed superior accuracy (R²=0.91, MAE = 6,950 kg), capturing complex ecological and management interactions. GIS spatial analysis identified significant yield clusters in intensive shrimp farms with a 12.5% compound annual growth rate, while natural water bodies and the Sundarbans exhibited declining productivity due to habitat degradation and salinity changes. PCA and ANOVA confirmed significant yield differences among aquatic environments, highlighting intensive farming benefits. ARIMA forecasting predicted general trends but was less accurate during anomalies. These results emphasize the need for targeted infrastructure, sustainable practices, and data-driven policies to improve resilience and productivity in Bangladesh’s shrimp aquaculture sector.