Using decision tree-based classification rules to improve land device measurement accuracy at a railway marshaling yard

Abstract

The paper discusses main paths to improve the effectiveness of railway intellectual marshalling yard monitoring and control systems and its land devices. It shows which environment conditions and other factors affect the accuracy of land device measurements particularly park track Occupancy Monitor Block which measures the length of unoccupied track portion. The paper shows that the measurements could be corrected by means of classification algorithms. It is suggested to use hierarchic decision trees as a classification rule. It discusses a method to improve classical decision trees building algorithms based on bootstrap method ideas. The suggested method improves accuracy of built decision trees.