Ocean wave prediction using Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) in Tuban Regency for fisherman safety

Abstract

The fishing industry has a large role in the Indonesian economy, with potential profits in 2020 of around US$ 1.338 billion. Tuban Regency is one of the regions in East Java that contributes to the fisheries sector. Fisheries relate to the work of fishermen. Accidents in shipping are still a major concern. One of the natural factors that influence shipping accidents is the height of the waves. Fisherman safety regulations have been established by the Ministry of Maritime Affairs and Fisheries and the Meteorology, Climatology and Geophysics Agency. Apart from regulations, the results of wave height predictions using the Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost) methods can help fishermen determine shipping departures, thereby reducing the risk of accidents. In this study, the Grid Search hyperparameter tuning process was used for both methods which were carried out on four location coordinates. Based on the analysis results, LSTM is superior in predicting wave height for the next 30 days because it can predict wave height at all three locations, with results at the first location (RMSE 0.045; MAE 0.029; MAPE 8.671 %), second location (RMSE 0.051; MAE 0.035; MAPE 10.64 %), and third location (RMSE 0.044; MAE 0.027; MAPE 7.773 %), while XGBoost only has the best value at fourth location (RMSE 0.040; MAE 0.025; MAPE 7.286 %).

• •
  Hyperparameter tuning with gridsearch is used in LSTM and XGBoost to obtain optimal accuracy
• •
  LSTM outperforms in three locations, while XGBoost outperforms in the fourth location.
• •
  Advanced prediction techniques such as LSTM and XGBoost improve fishermen's safety by providing accurate wave height estimates, thereby reducing the possibility of shipping accidents.