Applying the Ensemble and Metaheuristic Algorithm to Predict the Flexural Characteristics of Ice

Area of Science:

• Geophysics and climate science
• Materials science and engineering
• Computational intelligence

Background:

• Global warming and climate change pose significant threats to ice structures in polar and cold regions.
• Accurate estimation of ice flexural properties is essential for assessing the stability of these structures.
• The complex nature of ice failure and the influence of multiple factors necessitate advanced predictive models.

Purpose of the Study:

• To develop and evaluate data-driven artificial intelligence models for predicting ice flexural strength.
• To compare the performance of different machine learning algorithms, including ensemble methods.
• To identify key factors influencing ice flexural properties.

Main Methods:

• Development of artificial intelligence models: Classification and Regression Tree (CART), AdaBoost, and Random Forest.
• Optimization of model parameters using the Elitist Ant System (EAS).
• Evaluation of model accuracy and generalization using R-squared values and feature-importance analysis.

Main Results:

• The Elitist Ant System (EAS) rapidly optimized model parameters within ten iterations.
• Ensemble models demonstrated superior prediction accuracy and generalization compared to single CART models.
• AdaBoost achieved the highest prediction performance with an R-squared value of 0.736.
• Testing method and specimen geometry were identified as the most influential factors on flexural property predictions.

Conclusions:

• The proposed ensemble-metaheuristic framework offers an efficient approach for predicting ice mechanical behavior.
• These models provide valuable support for the stability assessment of ice structures under evolving climatic conditions.
• Careful control of experimental conditions is critical for reliable ice property estimations.