Journal of Agricultural Big Data ›› 2021, Vol. 3 ›› Issue (3): 55-61.doi: 10.19788/j.issn.2096-6369.210306

Previous Articles     Next Articles

Data Mining for Fishing Vessel Purchase Based on Gradient Boosting Decision Tree Algorithm

Yide Li1(), Feng Lu1,2(), Yong Zhu1, Shuo Xu1,2, Lu Sun1   

  1. 1.Institute of Fisheries Engineering, Chinese Academy of Fishery Sciences, Beijing 100141, China
    2.Pilot National Laboratory for Marine Science and Technology (Qingdao), Qingdao 266237, China
  • Received:2021-05-11 Online:2021-09-26 Published:2021-12-22
  • Contact: Feng Lu;


The purchase of a fishing vessel is a significant and complex process in the daily management of marine fishing fleets, and it yields the largest amount of data in all fishing vessel management operations. Through processing and analysis of the historical purchase data of fishing vessels, the potential decisive factors related to the purchase of fishing vessels can be found. This is significant to the protection of fishermen's economic interests and the development of fishing vessel management policies. We extracted and numerically processed the historical purchase data of fishing vessels from January 2018 to July 2020 using the physical and purchase data of fishing vessels in the Chinese Fishery Law Enforcement Command System (CFLECS) and taking Zhejiang Province as a typical case. The gradient boosting iterative decision tree (GBDT) algorithm was used to iterate the classifier regularly. We produced the results of feature classification and training set, and these were used to generate single decision tree and multiple decision tree models. We calculated the weight of the basic parameters of fishing vessels, such as length, material, and fishing type, to predict the potential possibility of fishing vessel transactions and to analyze the tendencies of fishermen when purchasing fishing vessels. The results indicate that age, length, trawler, and stow net are the principal determinants of fishing vessel transactions. The trawler and stow net vessel can only be obtained through the fishing vessel transaction. Thus, when the fishing vessel types are different, there is a great difference in the possibility of their being purchased. By comparing the loss functions of various features, we can find that the loss values of features with 20 years of age and ship length are more than 15% smaller than the loss values of other features, which means that the classification recognition rate calculated with selected features is higher. Consequently, quantitative analysis for fishermen's propensity to purchase fishing vessels can maximize fishermen's economic interests, and it can also play an auxiliary role in the formulation of fishing vessel management policy.

Key words: fishing vessel transaction, GBDT algorithm, decision tree, data mining, big data in fishery, Gradient Boosting Decision Tree, fishing vessel management

CLC Number: 

  • S972.7