Application Of Data Classify Method Based On Rule In Rail Transport Information

Along with the rapid development of China Railway Infomationizing Construction (CRIC), the scale of data produced by railway transport process expends extremely and the data types are exceedingly complicated, which bring new challenges to managers of rail system. However, current Railway Transport Manage Information System (RTMIS), which only provide some ordinary query and statistics functions and without the capability for real-time analysis and prediction on railway transport information, can not meet the needs of railway transport practice. Therefore, how to organize and utilize the mass railway transport data in an efficient way to discover the intrinsic relations among them and then, provide more accurate and direct instruct to railway transport management, becomes an important task to be solved in CRIC and is also the research purpose of this paper. In this paper, we introduce the progress of data mining and descript some general contents about statistics based classification (SBC). By analyzing two methods based on Naive Bayesian and Support Vector Machine (SVM), we investigate chief defects occurred in their applications to the classification of railway transport data. The result shows that, when there are serious overlaps among classes, the precision of classifier will descend. Especially, in hierarchical classification, the features of different subclasses is overlapped, the classification accuracy on subclasses is damaged even though father-classes are classified correctly, which can impede future classification in the next layer and finally cause the result of global classification becomes invalid. To address the above problems of statistical classification methods, we propose a novel rule based classifier for railway transport information. In this method, by establishing railway transport professional rules to design a rule based classifier, a robust classification model is build. Furthermore, we carry out a test by applying this model to RTMIS and get satisfied result. Finally, we make a comparative analysis of different methods. The result indicates that, due to the special characters (i.e. strong periodicity and seasonality), to obtain an excellent classification result, one must incorporated expert rules into SBC model and complements each other. Besides, the proposed rule based classification model is proved to possess outstanding generalization and extensive ability compared to the statistic based ones and therefore has a perspective future in railway transport information management application.