4.4 Limitations
(1) A total of 19 literatures were included in this study, all of which were in English, and only Su et al., Li et al., Xu et al., Tao et al., Guo et al., Wei et al. performed baseline Feature description. Xu et al., Shao et al., Deng et al. did not describe the blinding of performance bias. And the vast majority of literature, including Hu et al., reported on the existence of incomplete outcome data. (2) In the Egger’s test for the main indicators, although the results did not report the existence of publication bias, many hidden results will still affect the results of meta-analysis. (3) There is heterogeneity in the results of meta-analysis. After exploring the heterogeneity by means of subgroup analysis, the heterogeneity of some subgroups has been improved, but the rest of the subgroups still show some heterogeneity. The sources of these heterogeneities are very extensive, including but not limited to the experimental habits of researchers, laboratory light temperature environment, geographical latitude and magnetic field, etc. (4) Although the interpretability of the machine learning model has reached a high quality, the lack of sample data may still cause some trouble to the feature screening results.