Data-Driven Fault Text Classification of Urban Rail Transit Vehicle On-Board Signal System

In response to the problem of imbalanced textual records of urban rail transit faults, this paper proposes a data-driven method for automatically classifying urban rail transit vehicle on-board signal system fault texts. By leveraging the vehicle on-board signal system fault logs of urban rail transit, the pkuseg domain segmenter was trained to identify out-of-vocabulary words and perform Chinese word segmentation. Term Frequency-Inverse Document Frequency (TF-IDF) was employed for feature extraction, transforming the failure texts into word vectors. Adaptive Synthetic Sampling (ADASYN) was utilized for data augmentation to balance the minority class samples. Finally, the Support Vector Machine (SVM) algorithm was applied for fault text classification. Through the analysis of vehicle on-board signal system fault logs from a certain urban rail line spanning from 2016 to 2021, experimental results demonstrate that the proposed model can enhance word segmentation and fault classification performance in the field of rail transit.