PSI - Issue 84

Prajwal Giri et al. / Procedia Structural Integrity 84 (2026) 1119–1126 1121 class predictions that are combined using a weighted soft voting strategy. The ensemble probability for class c is computed as ( ) = ∑ = 1 ( ) , where ( ) denotes the probability assigned to class c by ℎ classifier, is its normalized weight based on validation performance, and N is the number of base learners, with ∑ = 1 =1 . The final predicted class label, denoted by ̂ , is obtained as ̂ = ( ) . In parallel, a hard voting strategy is also employed, where each base learner provides a categorical class label and the ensemble decision, is determined by majority rule as ̂ = [ 1 , 2 . . . . . ] , with representing the predicted label from the ℎ classifier. By integrating both voting mechanisms, the proposed ensemble framework reduces individual model bias and enhances robustness, generalization, and overall classification accuracy.

Fig. 1. Workflow of the proposed weighted ensemble learning framework.

3. Description of the bridge deck The experimental bridge specimen consists of three full-scale prestressed concrete I-shaped girders integrated with a 20 thick cast-in-situ reinforced concrete slab to form a composite deck system.

Fig. 2. (a) Overview of the bridge deck; (b) longitudinal profile of the bridge model; (c) cross-sectional view including the deck slab; and (d) beam profile illustrating the arrangement of post-tensioned and prestressed cables (dimensions in cm).