A Novel Ensemble Approach to Multi-label Classification for Electric Power Fault Diagnosis

State Grid Corporation of China has accumulated a large number of electric power fault textual data based on ICT customer services. Effectively classifying and analyzing these texts can provide important clues for resolving new faults, and therefore helps customer service staffs provide more accurate fault diagnosis solutions. Because an occurred fault is possibly related to multiple classification labels, it is challenging to effectively classify the faults. In this paper, we present an ensemble learning based multi-label classification approach to analyzing electric power fault text data. Firstly, the power fault report data is pre-processed by word segmentation and stop word removal according to the structure of fault data. Each of fault text is represented as a TF-IDF vector. Then, we combine Binary Relevance with the Gradient Boosting ensemble learning algorithm for multi-label classification of fault texts. At last, the related experiments were made, and the experimental results show that our method is better than the traditional approaches such as Binary Relevance based on Logistic Regression and ML-KNN for fault text classification.