A Study of Measurement Modeling of Decision Trees in Machine Learning Processes

Abstract

Accompanied by the rapid development of economy and science and technology, the ordinary measurement model with a single method of parameter determination and accuracy is not guaranteed, which has made it difficult to adapt to the measurement needs of complex data in industrial engineering and other systems. This study proposes a measurement model for complex data through the optimization of decision trees in the process of machine learning. Firstly, the gradient-boosting-based decision tree measurement model (GBDT) is constructed by analyzing the decision tree model, and then the model is solved. At the same time, latent variables were included in the model, SEM described the reflection relationship of explicit variables to latent variables, and the GBDT optimization model, including latent variables, was constructed by using the results of the model measurement, including latent variables. Then, for the measurement of multivariate data, the fusion convolutional network was used for image data feature extraction, and the combined measurement model with multi-source data fusion (MDF-DTFEE) was constructed on the basis of the decision tree measurement model. In the empirical analysis of the measurement model, the predicted and actual values of the model training were fitted between 4~60 mg/L and 5~45 ml/L, respectively, and its R² on the training set and test set were 0.948 and 0.886, respectively, with the RMSE lower than 1.2, and none of the MAPE exceeded 0.2. The practical application always had an error range of 1 mg/L, which is in line with the requirements. It fulfills the practical application requirements, demonstrates the practical value of the measurement model in this paper, and provides a useful solution for measuring complex data.