Dietary intake of micronutrients are predictor of premenstrual syndrome, a machine learning method

Abstract

Background & aims: Premenstrual syndrome (PMS) is a common disorder affecting 30-40% of women of reproductive age. Many modifiable risk factors associated with PMS involve nutrition and poor eating habits. This study aims to explore the correlation between micronutrients and PMS in a group of Iranian women and to build a predictor model showing the PMS using nutritional and anthropometric variables.

Methods: This is a cross-sectional study which was conducted on 223 females in Iran. Anthropometric indices were measured, including Body Mass Index (BMI) and skinfold thickness. Machine learning methods were used to assess participants' dietary intakes, Food Frequency Questionnaire (FFQ) and analyze the data.

Results: After applying different variable selection techniques, we have created machine learning models such as KNN. KNN achieved 80.3% accuracy rate and 76.3% F1 score indicating that our model is a curate and valid proof to show a strong relationship between input variables (sodium intake, Skin fold thickness of suprailiac, irregular menstruation, total calorie intake, total fiber intake, trans fatty acids, painful menstruation (dysmenorrhea), total sugar intake, total fat intake, and biotin) and the output variable (PMS). We sorted these effective variables based on their 'Shapley values' and figured out that Na intake, suprailiac skinfold thickness, biotin intake, total fat intake and total sugar intake have a major impact on having PMS.

Conclusions: Dietary intake and anthropometric measurements are highly associated with the occurrence of PMS, and in our model, these variables can predict PMS in women with a high accuracy rate.

Keywords: Machine learning; Micronutrients; Obesity; Premenstrual syndrome; Sodium.