An improved feature selection algorithm for cow subclinical mastitis

Abstract

Abstract Background Features of Dairy Herd Improvement dataset used for diagnosing subclinical mastitis in dairy cows contain important information about whether cows have subclinical mastitis, but they may also contain features that are unrelated or weakly related to the disease. The existence of these irrelevant or weakly correlated feature data increases the prediction time using machine learning models on the one hand, and reduces the accuracy of prediction on the other hand. In order to improve the prediction efficiency of the machine learning model of subclinical mastitis in dairy cows, feature selection is needed for Dairy Herd Improvement data. Results In this paper, an improved moth-flame feature selection algorithm was proposed and applied to the classification and prediction of subclinical mastitis in dairy cows. By introducing the dynamic adjustment strategy of flight direction and the position crossover strategy, the algorithm continuously generates new individuals while dynamically adjusting the flight direction of moths, which effectively avoids the feature selection algorithm falling into local optimum. By adaptively adjusting the number of flames, the population diversity is enhanced in the global exploration stage of the algorithm, and the premature convergence of the feature selection algorithm is avoided. Conclusion The improved feature selection algorithm and other comparison algorithms are experimentally verified on University of California Irvine (UCI) data sets and dairy cow subclinical mastitis disease data sets. The experimental results showed that the algorithm had better feature screening ability than other algorithms, and effectively improved the prediction performance of dairy cow recessive mastitis disease.