# Using Machine Learning to Identify the Dynamic Evolution Patterns of Negative Emotions in Perinatal Women: A Longitudinal Study in Southwest China

## TL;DR

This study tracks negative emotions in pregnant women over time and uses machine learning to identify high-risk groups.

## Contribution

The study introduces a machine learning-based model to predict high-risk negative emotional trajectories in perinatal women.

## Key findings

- Negative emotional trajectories were categorized into four distinct groups.
- The extreme gradient boosting model achieved the highest prediction performance score.
- Machine learning models show potential in identifying high-risk negative emotions in perinatal women.

## Abstract

Paying attention to the mental health of perinatal women is helpful in improving their quality of life. However, the existing research pays less attention to the heterogeneity of its negative emotional trajectory and the identification of high‐risk groups. This study recruited 860 perinatal women from four large hospitals in Chongqing from March 2018 to January 2019. They were followed up by structured questionnaires in the first trimester, second trimester, third trimester, and about 6 weeks after delivery. The growth mixture model was used to analyze the developmental trajectory of negative emotions, and six machine learning algorithms were used to establish a high‐risk negative emotion recognition model. The performance of the model was comprehensively evaluated by five performance indicators. The SHAP algorithm was used to explain the model. Negative emotional trajectories were divided into four categories: low‐stable anxiety group, gradually increasing high‐anxiety group, mild sustained depression group, and high‐progressive depression group. The extreme gradient boosting model performed best, with the highest prediction performance score (24 points). In summary, the negative emotional trajectory of perinatal women is dynamic and heterogeneous, and the prediction model based on machine learning may play an important role in identifying high‐risk negative emotions.

This study examined the dynamic evolution of negative emotions in perinatal women. By utilizing a growth mixture model on a longitudinal cohort of pregnant women in Southwest China, we accurately identified four distinct trajectories of negative emotional development. Additionally, we explored the potential of machine learning models to predict high‐risk negative emotions.