PLoS ONE (Jan 2023)

Comparing machine learning methods for predicting land development intensity.

  • Guanhai Gu,
  • Bin Wu,
  • Wenzhu Zhang,
  • Rucheng Lu,
  • Xiaoling Feng,
  • Wenhui Liao,
  • Caiping Pang,
  • Shengquan Lu

DOI
https://doi.org/10.1371/journal.pone.0282476
Journal volume & issue
Vol. 18, no. 4
p. e0282476

Abstract

Read online

Land development intensity is a comprehensive indicator to measure the degree of saving and intensive land construction and economic production activities. It is also the result of the joint action of natural, social, economic, and ecological elements in land development and utilization. Scientific prediction of land development intensity has particular reference significance for future regional development planning and the formulation of reasonable land use policies. Based on the inter-provincial land development intensity and its influencing factors in China, this study applied four algorithms, XGBoost, random forest model, support vector machine, and decision tree, to simulate and predict the land development intensity, and then compared the prediction accuracy of the four algorithms, and also carried out hyperparameter adjustment and prediction accuracy verification. The results show that the model with the best prediction performance among the four algorithms is XGBoost, and its R2 and MSE between predicted and valid values are 95.66% and 0.16, respectively, which are higher than the other three models. During the training process, the learning curve of the XGBoost model exhibited low fluctuation and fast fitting. Hyperparameter tuning is crucial to exploit the model's potential. The XGBoost model has the best prediction performance with the best hyperparameter combination of max_depth:19, learning_rate: 0.47, and n_estimatiors:84. This study provides some reference significance for the simulation of land development and utilization dynamics.