# Comparative analysis of clinical feature–based machine learning models for predicting myofascial pelvic pain syndrome: a single-center retrospective study

## TL;DR

This study compares machine learning models to predict myofascial pelvic pain syndrome in women, finding random forest as the most accurate method for early diagnosis.

## Contribution

The study introduces a machine learning-based prediction model for MPPS using clinical features, specifically optimized for Chinese women.

## Key findings

- Random forest achieved the highest accuracy (0.91) and AUC (0.956) for predicting MPPS.
- XGBoost and LightGBM also showed strong performance with AUCs above 0.95.
- Machine learning models can support early diagnosis and personalized treatment for MPPS.

## Abstract

Myofascial pelvic pain syndrome (MPPS) is a common but often underdiagnosed cause of chronic pelvic pain in women, significantly affecting quality of life. Early and accurate identification of patients at risk is essential for improving treatment outcomes and reducing the clinical burden.

This study aimed to develop an effective machine learning-based prediction model for MPPS among Chinese women to assist in early diagnosis and personalized treatment.

A total of 1,136 women diagnosed with MPPS and 1,448 healthy women who underwent pelvic floor screening during the same period were included, yielding 2,584 samples. Six machine learning algorithms—logistic regression, support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and adaptive boosting (AdaBoost)—were trained using 5-fold cross-validation and grid search. Model performance was evaluated using the confusion matrix, precision, recall, F1 score, overall accuracy, and receiver operating characteristic (ROC) curves.

The accuracies of the six models were 0.77, 0.80, 0.91, 0.89, 0.88, and 0.81, respectively. The average area under the ROC curves (AUCs) were 0.670, 0.672, 0.956, 0.951, 0.952, and 0.836, respectively. Among the models, RF achieved the best performance for predicting MPPS, while XGBoost and LightGBM performed slightly lower, with all three models having AUCs above 0.95.

Machine learning models, particularly the random forest algorithm, demonstrated strong potential for accurately predicting MPPS, supporting early diagnosis and enabling personalized clinical decision-making for women affected by this condition.