Improved random forest classification model combined with C5.0 algorithm for vegetation feature analysis in...

Area of Science:

• Remote Sensing
• Machine Learning
• Ecology

Background:

• Traditional random forest algorithms struggle with high-dimensional, noisy satellite data for non-agricultural vegetation classification.
• Challenges include high computational complexity and suboptimal classification performance.

Purpose of the Study:

• To propose an enhanced random forest algorithm for improved classification of non-agricultural vegetation using satellite data.
• To address the limitations of traditional methods in high-dimensional and noisy datasets.

Main Methods:

• Developed an enhanced random forest algorithm integrating the C5.0 algorithm for feature selection.
• Employed an ensemble feature method based on the bagging concept to improve feature selection and model diversity.
• Utilized the enhanced vegetation index (EVI) for vegetation coverage estimation.

Main Results:

• The object-oriented random forest model achieved 94.02% accuracy on an aerial imagery dataset.
• The proposed algorithm reached an average accuracy of 90.20% in identifying non-agricultural vegetation features.
• Outperformed other models like BERT, FastText, and CNN with accuracies ranging from 84.41% to 88.33%.

Conclusions:

• The enhanced random forest algorithm effectively classifies non-agricultural vegetation, offering improved accuracy and efficiency.
• Findings provide scientific evidence supporting agricultural ecosystem protection and biodiversity restoration efforts.
• The C5.0 algorithm and EVI contribute to robust feature selection and vegetation coverage estimation.