Sibanda Khulekani

A COMPARATIVE ANALYSIS OF LARGE LANGUAGE MODELS AND TRADITIONAL MACHINE LEARNING MODELS FOR PREDICTION OF PREECLAMPSIA

Ncube Mthandazo Josephat Junior¹; Sibanda Khulekani²

Preeclampsia remains a significant cause of maternal complications and death, therefore early and accurate prediction is critical to reducing complications. Traditional machine learning models rely on structured data such as age, BMI, and blood pressure, while large language models (LLMs) offer the advantage of processing both structured and unstructured data, such as clinical notes. This study developed and compared the performance of traditional machine learning models (Random Forest, XGBoost) and an LLM (BERT) in predicting preeclampsia, using maternal health data from Mpilo Central Hospital and Kaggle. This study follows the CRISP-DM framework for data preprocessing, model training, and evaluation and utilizes a machine learning-based quantitative research methodology. Data preprocessing using Pandas and NumPy and feature selection using SelectKBest preceded model training and evaluation. Random Forest and XGBoost were implemented with scikit-learn, while BERT used Hugging Face Transformers. Evaluation metrics included accuracy, precision, recall, and F1-score. Result demonstrated BERT as the top performer with an accuracy of 92%, outperforming XGBoost (90%) and Random Forest (88%). LLMs improved predictive performance by leveraging unstructured data alongside structured features. Results show that predictive performance on clinical risk prediction tasks can be improved using LLMs by leveraging unstructured data, and they are promising for healthcare decision-making improvement.