Intelligent identification of coal miners with fatty liver under a cascade reduction strategy

Abstract

Precise and efficient assisted diagnosis of fatty liver disease in coal miners is directly related to the development of occupational health prevention and control efforts in the coal mining industry. We proposed a cascade reduction strategy based on neighbourhood component analysis (NCA) joined with expectation maximization and principal component analysis (EM-PCA) to address the shortcomings of traditional manual diagnostic methods such as low efficiency, missed diagnosis, misdiagnosis, and insufficient mining of necessary hidden information. We also developed a classification model under intelligent optimization algorithms for the identification of fatty liver in coal miners. First, the performance of different classification algorithms is compared to determine the basic classifier of extreme learning machine (ELM) for identifying fatty livers in coal miners. Then, four new continuous indicators are added to restructure the raw data. The NCA is used to remove redundant interference information that affects the model complexity and to screen out nine important feature parameters. Finally, the EM-PCA is synergized with the ELM of intelligent optimization algorithm by slime mould algorithm (SMA-ELM) is applied to further simplify the rest variable and obtain the optimal model with data of seven defining features. Meanwhile, the average accuracy, F1-score, Matthews correlation coefficient and time cost of the relatively excellent model were 95 %, 0.9652, 0.8781 and 1.5692 s. Experimental results show that the proposed cascade reduction strategy achieves accurate identification of fatty liver in coal miners with fewer features. The conclusions of this study can serve as a reference for early intelligent screening, intelligent health management and intelligent assisted diagnosis of occupational health in coal miners.