Tree-Based Pipeline Optimization-Based Automated-Machine Learning Model for Performance Prediction of Materials and Structures: Case Studies and UI Design

Abstract

Machine learning (ML) methods have become increasingly prominent for predicting material and structural performance in civil engineering. However, these methods often require repetitive iterations and optimizations by professionals to obtain an optimal model, which are time-consuming and challenging for nonexpert users. In this paper, we propose an automated ML (Auto-ML) model using the tree-based pipeline optimization tool (TPOT) to address these limitations and streamline the performance prediction process. TPOT leverages genetic programming to optimize various ML models, including DT, RF, GBDT, LightGBM, and XGBoost, and to search possible models that fits a particular dataset, which cuts the most tedious parts of ML. To demonstrate the effectiveness of TPOT-based Auto-ML, two case studies are presented by using TPOT-based Auto-ML algorithms to construct prediction models for compressive strength of recycled micropowder mortar, and punching shear bearing capacity/failure mode of RC slab-column joints. To explain the “black box” of Auto-ML, Shapley Additive Explanation (SHAP) is introduced to interpret the best predictive models and rank the importance of influencing factors, providing a basis for material and structural design. Finally, a user interface (UI) for engineering applications is developed which enables end-to-end automation from data preprocessing to predictive results presentation.