Active Learning for Multi-way Sensitivity Analysis with Application to Disease Screening Modeling.

Area of Science:

• Computational modeling
• Data science
• Biostatistics

Background:

• Sensitivity analysis is crucial for assessing model confidence but is computationally intensive.
• Evaluating numerous parameter combinations requires significant time and resources.
• Identifying representative parameter subsets can mitigate computational burden.

Purpose of the Study:

• Investigate machine learning (ML) for accelerating sensitivity analysis.
• Apply feature selection to determine quantitative model parameter importance.
• Evaluate active learning strategies for optimizing ML model construction in sensitivity analysis.

Main Methods:

• Utilized ML algorithms, including Random Forests and XGBoost, for prediction tasks.
• Employed feature selection to identify influential model parameters.
• Implemented active learning strategies to guide the selection of informative parameter combinations.

Main Results:

• Ensemble methods (Random Forests, XGBoost) demonstrated superior performance in prediction tasks.
• Feature selection identified relative importance of quantitative model parameters.
• Active learning significantly reduced the number of required model runs for high-performance prediction models.

Conclusions:

• ML-based approaches, particularly ensemble methods, effectively accelerate sensitivity analysis.
• Active learning enhances efficiency by selecting optimal parameter combinations for model training.
• These methods offer substantial speed-ups for computationally demanding sensitivity analyses in disease screening models.