Stacked hybrid model: Multi-layer perceptron and logistic regression with meta-learning for cesarean section classification

Hybrid model for C-section prediction

Abstract

Background/Aim: This study aims to develop an interpretable and practical decision support method for early prediction of the need for cesarean delivery. Although machine learning and deep learning models are prevalent in the literature, their generalization capabilities are often restricted, especially when utilizing small clinical datasets. This limitation underscores the necessity for robust, transparent, and well-regularized models in medical decision-making processes.

Methods: The study proposed a stacking-based hybrid model, which combines the strengths of both classical and modern techniques. The data were normalized using StandardScaler, and feature selection involved principal component analysis (PCA) and SelectKBest to capture global and target-relevant patterns. In the classification phase, two parallel learners – a regularized multi-layer perceptron (MLP) and logistic regression – were used, followed by a random forest meta-learner.

Results: The experimental analysis demonstrated that the proposed model achieved an average accuracy of 96.43% under stratified 5-fold cross-validation. Although this result surpassed the performance of other baseline models within the dataset, it should be regarded as preliminary due to the limited sample size.

Conclusion: The findings indicate that the proposed hybrid approach has potential as a promising direction for future clinical decision support research. Nonetheless, additional validation using larger and more diverse datasets is necessary to adequately assess its generalizability and practical utility.