.png){kind=icon}
.png){kind=icon}
Return
Complex product quality prediction method based on an improved light gradient boosting machine
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-024-06112-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Quality prediction, as a means of identifying potential quality issues in products, plays a crucial role in increasing the level of quality control within enterprises. The data from the process of manufacturing complex products exhibit characteristics of high dimensionality, strong correlation, and imbalance, which pose certain challenges to achieving accurate quality prediction for complex products, especially in intervals with sparse sample distributions. To improve the accuracy of quality prediction for complex products, this paper proposes a complex product quality prediction model based on cost-sensitive learning and gradient boosting decision trees (GBDTs). Initially, eXtreme gradient boosting (XGBoost) is employed to select the optimal feature subset from the original high-dimensional data. A mapping relationship between the manufacturing process data and quality characteristic values is subsequently established on the basis of a light gradient boosting machine (lightGBM) model. On this basis, cost-sensitive learning is introduced, and a new loss function named DenseMSE is designed for the lightGBM model, establishing a quality prediction model based on DenseMSE-lightGBM. The experimental results demonstrate that the proposed quality prediction model has improved the prediction accuracy in intervals with sparse samples. Moreover, the accuracy of the quality prediction model based on DenseMSE-lightGBM surpasses that of other mainstream prediction models, providing meaningful guidance for achieving more accurate quality prediction for complex products.
Keywords:
Quality prediction
Feature selection
Gradient boosting decision trees
Cost-sensitive learning
Imbalanced data regression
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W