Research and Optimization of Feature Selection Method for TMS Classification Learning

Due to the expansion of data scale, the data in Transportation Management System (TMS) of the State Grid has the features of distributed storage, more data noise and larger data dimension, making data analysis extremely difficult. Therefore, data sorting and feature selection should be carried out on the basis of the original data for subsequent learning tasks. The optimization method of feature selection for TMS classification learning is able to effectively select a large number of features, thus improving the accuracy of classification learning. In order to train a feature selection model suitable for TMS, this paper studies the method of applying random forest model integrated by multiple decision trees to TMS feature selection. By comparing the number of decision trees, the criteria of feature classification, the maximum number of features in candidate subset of feature classification, and the change of model accuracy after feature rearrangement with key algorithms, a feature selection method most suitable for TMS data is given and verified by experiments. Experimental results show that the proposed optimization method of feature selection lays a good foundation for future learning tasks.