Machine Learning and Deep Learning Techniques for Predictive Modeling of Marine Ecosystem – A case of Flic en Flac Region, Mauritius

The uncertainty and adverse impact of climate change on marine ecosystems is highly evident through diminishing numbers of fish assemblages and hard corals within lagoons. Hence, modeling and predicting the future climatic conditions and existence of flora and fauna is quite challenging. This research carried out in Flic en Flac lagoon of Mauritius, aimed to investigate the relationship between the pertinent independent variables / key predictors such as Sea Surface Temperature, pH, Ocean Salinity, Chemical Oxygen Demand, Nitrate-Nitrogen, Phosphate and dependent variables such as the hard corals, fish assemblages, algae and abiotic factors. Researchers used these historic data and evaluated predictive modelling using various machine learning and deep learning models, including Random Forest Classifier (RF), Bagging Regressor (BR) AdaBoost Regressor (ABR), Linear Regression (LR), Time Series forecast using Seasonal Auto Regressive Integrated Moving Average (SARIMAX), and Long Short-Term Memory (LSTM). The in-depth data analysis revealed as first finding, a strong association between pH, sea surface temperature and marine community structure. Secondly, by comparing the model evaluation metrics of the considered algorithms, the study also confirmed that SARIMAX has better predictive accuracy over other models. This research concludes that proposed predictive model which is a Combined Multiple Regression Algorithms consisting of Linear Regression, Random Forest Regressor, Support Vectors Regression, Multi-Layer Perceptron Neutral Network Regression, is taken as voting ensemble regressor and is found reliable for predictive modeling as the prediction metrics recommend their combined use. Research findings further established that the hybrid approaches, such as training and evaluating SARIMAX and LSTM on pre-clustered data, may enhance predictive quality.