Models

LighGBM

LightGBM uses leaf-wise tree growth strategy, which effectively finds the leaves with the highest branching gain each time from all the leaves, and then goes through the branching cycle. Therefore, it can reduce more errors and obtain better precision with the same number of times of segmentation. A maximum depth limit is set to prevent overfitting while ensuring high efficiency. Leaf-wise trees can sometimes overfit especially with smaller datasets. Limiting the tree depth can help to avoid overfitting.

The result of the first attempt in using the LightGBM model: Overfitting

Reasons: The smallness of the dataset Large number of model parameters