Scholarlink

<p>Marine ecosystems are declining, due to harmful effects of climate change and hence marine ecosystem prediction is the need of the hour to work towards ocean sustainability. Considering this context, this research performed stationarity of data, and pre-clustering of data obtained for Flic en Flac lagoon of Mauritius using clustering algorithms. Based on their intrinsic evaluation metrics this research determined that Self Organising Map (SOM) is the appropriate clustering algorithm. Further, this research predicted mean population density of Hard Corals (HC), and fish assemblages which are subjected to abiotic factors that include Sea surface temperature, Practical Salinity, pH, and Chemical Oxygen Demand using variants of deep learning algorithms namely Recurrent Neural Networks (RNN), Long Short-Term Memory Networks (LSTM), and Convolution Neural Networks (CNN). Comparative analyses of these predictive algorithms’ effectiveness were done by determining their performance metrics. The predicted results of DeepRNN showed a reduction in mean absolute error of 48.87% than LSTM and 91.78% than CNN. The root mean square error of DeepRNN showed reduction of 36.14% than LSTM and 85.01% than CNN. Thus, this research proposes that hybrid SOM clustered DeepRNN predictive model has better prediction efficiency than other variants which was further validated by external validation. Hence, for the region under consideration, this research informs that the mean population density of HC may decrease by 1.48%, fish assemblages: Pomacentridae by 9.45%, Chaetodontidae by 10.13%, Labridae by 13.65% in the year 2025 when compared to 2022. As way forward, Hyper Parameter Optimization can improve prediction accuracy of DeepRNN.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative analysis of RNN, LSTM and CNN algorithms for marine data prediction

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Agnes Nalini Vincent,
  • K. Sakthidasan,
  • Mohammad Sameer Sunhaloo,
  • J. Christopher Clement,
  • Nassirah Laloo,
  • Uhoze Bagurubumwe

摘要

Marine ecosystems are declining, due to harmful effects of climate change and hence marine ecosystem prediction is the need of the hour to work towards ocean sustainability. Considering this context, this research performed stationarity of data, and pre-clustering of data obtained for Flic en Flac lagoon of Mauritius using clustering algorithms. Based on their intrinsic evaluation metrics this research determined that Self Organising Map (SOM) is the appropriate clustering algorithm. Further, this research predicted mean population density of Hard Corals (HC), and fish assemblages which are subjected to abiotic factors that include Sea surface temperature, Practical Salinity, pH, and Chemical Oxygen Demand using variants of deep learning algorithms namely Recurrent Neural Networks (RNN), Long Short-Term Memory Networks (LSTM), and Convolution Neural Networks (CNN). Comparative analyses of these predictive algorithms’ effectiveness were done by determining their performance metrics. The predicted results of DeepRNN showed a reduction in mean absolute error of 48.87% than LSTM and 91.78% than CNN. The root mean square error of DeepRNN showed reduction of 36.14% than LSTM and 85.01% than CNN. Thus, this research proposes that hybrid SOM clustered DeepRNN predictive model has better prediction efficiency than other variants which was further validated by external validation. Hence, for the region under consideration, this research informs that the mean population density of HC may decrease by 1.48%, fish assemblages: Pomacentridae by 9.45%, Chaetodontidae by 10.13%, Labridae by 13.65% in the year 2025 when compared to 2022. As way forward, Hyper Parameter Optimization can improve prediction accuracy of DeepRNN.