GBWOEM: A Gradient-Based Weight Optimization Model for Improved Predictive Accuracy in Healthcare.

Area of Science:

• Machine Learning
• Ensemble Learning
• Healthcare Analytics

Background:

• Ensemble learning is vital for improving predictive accuracy in healthcare diagnostics and classification.
• Accurate predictions are critical as they directly impact patient outcomes.
• Ensemble models mitigate misclassification risks by integrating multiple machine learning models.

Purpose of the Study:

• Introduce the Gradient-Based Weight Optimized Ensemble Model (GBWOEM).
• Optimize weights of five base models (DTC, RFC, LR, MLP, KNN) for enhanced performance.
• Evaluate two GBWOEM variants (GBWOEM-R and GBWOEM-U) on diverse healthcare datasets.

Main Methods:

• Developed GBWOEM, an ensemble technique optimizing base model weights.
• Utilized five base models: Decision Tree Classifier (DTC), Random Forest Classifier (RFC), Logistic Regression (LR), Multi-Layer Perceptron (MLP), and K-Nearest Neighbours (KNN).
• Tested GBWOEM variants with random (GBWOEM-R) and uniform (GBWOEM-U) weight initialization on five healthcare datasets.

Main Results:

• Achieved 0.48-8.26% increase in test accuracy compared to traditional ensemble models (Adaboost, Catboost, GradientBoost, LightGBM, XGBoost).
• Demonstrated efficacy in handling imbalanced data through ROC-AUC analyses.
• Confirmed improved predictive consistency in healthcare applications.

Conclusions:

• GBWOEM enhances predictive accuracy and reliability in healthcare, particularly for imbalanced data.
• The model contributes to improved patient outcomes and diagnostic consistency.
• GBWOEM presents a robust solution for critical healthcare prediction tasks.