Hybrid Crow Search and ICA Optimized Regression Random Forest Model for Precise Preterm Pregnancy Risk Prediction Assessment

Abstract

This paper suggests a hybrid machine learning method, called Hybrid Crow-ICA with Regression Random Forest (HyCIRRF), to estimate the likelihood of having a preterm pregnancy based on maternal electronic health records (EHRs). It consists of a framework of Fast Independent Component Analysis (Fast-ICA) to extract features, CatBoost to rank the importance of features, and Crow Search Optimization (CSO) to select the best features. The last predictive ensemble is a hybrid of Regression Random Forest (RRF) and Logistic Regression (LR) with a soft-voting mechanism to enhance the productivity and interpretability. The model was tested on the Maternal Health Risk Dataset, and the model had an accuracy of 93%, a precision of 93%, a recall of 94, and an error rate of 7, which is the best balance between sensitivity and specificity. Moreover, the values of AUC greater than 0.95 represent great discrimination of low-, medium-, and high-risk groups. Compared to other baseline models, such as the Logistic Regression, Decision Tree, SVM, and XGBoost, HyCIRRF proved to be as accurate as those with explainability and computationally efficient simultaneously. These results indicate that this proposed model may be applied as a valid and understandable clinical decision support system in the early identification of high-risk pregnancies.