Multi-Objective Performance Optimization of Machine Learning Models in Healthcare.

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Optimizing machine learning (ML) for clinical diagnostics requires balancing sensitivity and specificity.
• Missed diagnoses (low sensitivity) and unnecessary procedures (low specificity) have significant clinical and economic impacts.

Purpose of the Study:

• To develop a multi-objective optimization framework (MOOF) for ML models in medical applications.
• To simultaneously optimize model parameters for accuracy, sensitivity, and specificity.

Main Methods:

• MOOF employs the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS).
• The framework optimizes parameters for Random Forest, Support Vector Machine, and Multilayer Perceptron algorithms.
• Performance was evaluated against multi-score grid search and single-objective optimization methods.

Main Results:

• MOOF demonstrated superior performance compared to traditional optimization techniques.
• The framework effectively provides optimal solutions that represent trade-offs between sensitivity, specificity, and accuracy.
• Optimized models showed potential for enhancing diagnostic precision.

Conclusions:

• Multi-objective optimization is crucial for developing precise ML models in medical informatics.
• MOOF offers a powerful approach to enhance ML model performance for clinical decision support.
• This methodology has the potential to improve patient care through more accurate diagnostics.