Interpretable Machine Learning Models for Identifying Premenstrual Syndrome and Related Factors in Young Adult Women

Abstract Background Premenstrual syndrome (PMS) is a public health problem with widespread impact, influenced by the multiple factors. Young adult women, as a high-risk group for PMS, can benefit from early identification of key factors related to PMS. This study aimed to construct a machine learning model for identifying PMS among young adult women and explore the complex relationships between associated factors and PMS. Method A secondary analysis was performed using cross-sectional data from 3447 young adult women. Using a score of 6 points on the Premenstrual Syndrome Scale (PSS) as the cutoff, all participants were divided into the PMS group and the non-PMS group. The dataset was randomized into a training set and a validation set in 75% and 25% proportions. Five machine learning algorithms were used to develop models and the output of performed-best model was interpreted using SHapley Additiveex Planations (SHAP). Results There were 1474 women in PMS group and 1973 women in non-PMS group. Among five machine learning models, the random forest model performed best, with an AUC of 0.782. Neuroticism was most strongly associated with PMS, followed by mindfulness, history of dysmenorrhea, emotional abuse, and use of analgesic. Conclusion The results suggested that the random forest model was an effective tool for identifying PMS among young adult women, and neuroticism could serve as a crucial predictive indicator. Furthermore, interventions such as mindfulness training, dysmenorrhea management, and addressing early trauma may hold significant potential in preventing and managing PMS.