Comparing machine learning methods for predicting land development intensity.

Area of Science:

• Environmental Science
• Geospatial Analysis
• Urban Planning

Background:

• Land development intensity (LDI) is a key metric for assessing land use efficiency and economic activity.
• Understanding LDI dynamics is vital for sustainable regional development and effective land use policy formulation.
• LDI is influenced by a complex interplay of natural, social, economic, and ecological factors.

Purpose of the Study:

• To simulate and predict land development intensity across Chinese provinces.
• To compare the predictive accuracy of four machine learning algorithms: XGBoost, random forest, support vector machine, and decision tree.
• To identify the optimal algorithm and hyperparameters for accurate LDI prediction.

Main Methods:

• Utilized inter-provincial land development intensity data and associated influencing factors.
• Applied and compared four machine learning algorithms: XGBoost, random forest, support vector machine, and decision tree.
• Performed hyperparameter tuning and accuracy verification for model optimization.

Main Results:

• XGBoost exhibited the highest prediction performance among the tested algorithms, achieving an R2 of 95.66% and an MSE of 0.16.
• The XGBoost model demonstrated a stable learning curve with rapid fitting during training.
• Optimal hyperparameters for XGBoost were identified as max_depth:19, learning_rate:0.47, and n_estimators:84.

Conclusions:

• XGBoost is the most effective algorithm for predicting land development intensity based on the study's data.
• Hyperparameter optimization significantly enhances the predictive power of machine learning models for LDI.
• The findings provide a valuable reference for simulating land development and utilization dynamics and informing policy.