A Comparative Evaluation of Machine Learning- AI Based Fish Catch Prediction for Ratnagiri Coastal Fisheries

Abstract

Marine fishing is a main source of income in Ratnagiri district, yet most decisions about when and where to fish still rest on personal experience. This paper reviews AI and machine learning work in fisheries, covering vessel tracking, temporal catch prediction, ML-based fishing prediction, data standardization, environmental data and AI-based monitoring. Most of these studies deal with a single task, one region or one data type, or stop before forecasting catch quantity. We also summarize a Ratnagiri project that links historical catch records with Google Earth Engine environmental variables for zone-wise prediction using XGBoost and district-wise prediction using Random Forest and LSTM. In the zone-wise experiment XGBoost gave an MAE of 14,212.93 t, an RMSE of 23,218.80 t and an R² of 0.46, and a walk-forward check of a gradient boosting model over 2018–2021 gave an MAE of 8,761.36 t and an R² of 0.51 with large errors in individual zones. The test sets are very small, so the results are indicative at best. Catch values for future years are model-based forecasts, not observed or guaranteed figures.

Keywords