An Approach to Recommend Fishing Location and Forecast Fish Production by Using Big Data Analysis and Distributed Deep Learning
Minh-Triet Thai, Thao-Ngan Chu-Ha, Tuan-Anh Vo, Trong-Hop Do · 2022 RIVF International Conference on Computing and Communication Technologies (RIVF) · 2022
In fishing industry, fish populations can move over a vast sea and thus boats often search for days or weeks before making a catch. Given the excessive CO2 emissions from vessels and rising fuel cost, it is important to optimize commercial fishing activities by reducing un-necessary searching period. This study proposes a hybrid recommendation system for predicting the best locations for catching specific fish species and experimented various deep learning based multi-variate time series models to forecast the gross weight of two important species of Norwegian fishery: haddock and mackerel. To ensure the practicality, both recommendation system models and time series forecasting models are trained and deployed using Apache Spark and BigDL, which are frameworks for big data processing and distributed deep learning training. The proposed hybrid recommendation model achieve good performance with RMSE of 0.4933. Deep learning based models are also shown to achieve high performance on forecasting fish gross weight. Through this study, it is revealed that datetime and environmental features can play important roles for building sustainable commercial fishing plan.