Modeling Dynamic Systems with Efficient Ensembles of Process-Based Models.

Area of Science:

• Environmental Science
• Machine Learning
• Computational Biology

Background:

• Ensemble models enhance predictive accuracy and robustness in machine learning.
• Process-based models utilize domain knowledge for dynamic systems analysis.
• Existing ensemble methods like bagging and boosting improve process-based models but increase computational cost.

Purpose of the Study:

• To develop an efficient method for learning ensembles of process-based models.
• To maintain accurate long-term predictive performance in dynamic systems.
• To reduce the computational time associated with ensemble learning.

Main Methods:

• Proposed a new method for learning ensembles by sampling domain-specific knowledge.
• Applied the method to automated predictive modeling in three lake ecosystems.
• Utilized a library of process-based knowledge for population dynamics modeling.

Main Results:

• The proposed method efficiently learns ensembles of process-based models.
• Ensembles generated using domain knowledge sampling show significantly improved population dynamics predictions.
• The new ensembles achieve comparable predictive performance to bagging and boosting but are substantially more efficient.

Conclusions:

• Sampling domain knowledge is an efficient strategy for learning process-based model ensembles.
• The proposed method offers a computationally efficient alternative for accurate long-term prediction in dynamic systems.
• This approach advances automated predictive modeling in ecological systems.