Research on the Prediction of Nonbreakeven Financial Products’ Yield of Commercial Banks Based on Machine Learning

Table 3

The parameter of the random forest model.

Parameters	Descriptions and settings
Task	Train
Objective	Model training target, which can be selected as a regression model or binary classification model
Boosing_type	Base learner, for gbdt
Metric	Metrics as a function of evaluating the optimal model
Leaning_rate	Learning rate, the default is 0.1, the smaller the value, the more accurate the learning
n_estimators	The number of iterations in the classifier
Num_leaves	The number of leaves in the tree, the indicator should be less than 2 to the power of max_depth
Max_depth	Tree depth
Feature_fraction	The feature sampling ratio for building the tree, which ranges from 0 to 1
Bagging_fraction	The sample sampling ratio for building the tree, the data range is 0 to 1
Max_bin	The maximum bin value, generally equal to the number of features
Min_data_in_leaf	The minimum number of samples for each leaf node, when the leaf node is smaller than this value, the tree will no longer be split
Min_gain_to_split	The tree stops splitting when the leaf node is smaller than this value