Ensemble Recognition Based on the Harmonic Information Gain Ratio for Unsafe Behaviors in Coal Mines
| Autor: | Yi-nan Guo, Cheng Jian, Shijie Wang, Jiao Botao |
|---|---|
| Rok vydání: | 2021 |
| Předmět: | |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783030788100 ICSI (2) |
| Popis: | More than 90% accidents occurred in coal mine are caused by unsafe behaviors of human. How to effectively identify unsafe behaviors and decrease the possibility of their occurrence is the fundamental of avoiding accidents. However, the number of unsafe behaviors is far less than that of safe ones in a behavior dataset of coal mine. Serious imbalance has a negative impact on recognition efficiency and accuracy. To address the problem, the harmonic information gain ratio is defined by introducing the degree of imbalance into traditional information gain, and the corresponding feature selection method is presented. By integrating it into Underbagging, a novel ensemble recognition based on the harmonic information gain ratio for unsafe behaviors is presented, with the purpose of avoiding information loss caused by feature reduction and guaranteeing recognition accuracy. Based on a sub-dataset obtained by undersampling, the optimal features subset is selected by the proposed feature selection method, and employed to train a base classifier built by support vector machine. The weighted sum of all base classifiers output forms final recognition result. Each weight is calculated from the corresponding harmonic information gain ratio. Experimental results on UCI dataset and a behavior dataset for a particular coal mine indicate that the proposed ensemble recognition method outperforms the others, especially for a dataset with high imbalance ratio. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|