Prediction of Reinforced Concrete Beam-Column Joint Capacity Considering Shear Stress-Strain Using Feedback Propagation in MATLAB/ Bea Marie L. Andrade, Carla Jane C. Evangelista, Caitlyn A. Merelos, Kyla G. Remetio and Vincent A. Salvador.--

Bachelor's thesis

College of Engineering.-- Bachelor of Science in Civil Engineering: Technological University of the Philippines, Manila. 2024

Includes bibliographic references and index.

Artificial Neural Networks (ANNs) are widely applied in construction
engineering, particularly structural analysis and design. This study investigates the use of
ANNs for predicting beam-column joint capacity based on shear stress-strain, utilizing
MATLAB as the implementation platform. An ANN model was developed and trained to
enhance predictive accuracy and optimize structural design parameters. A dataset of 517
samples was compiled from 137 research studies, with 153 outliers identified and
removed using Cook's Distance Method in MATLAB, leaving 364 datasets for training.
The network was trained using the Levenberg-Marquardt (LM) algorithm, a widely used
iterative technique for nonlinear least-squares problems, combined with the TANSIG

activation function. The model achieved a mean squared error (MSE) of 0.480 and an R-
value of 0.94, with the optimal configuration being 36 layers. Validation was conducted

through simulations, weight, and bias computations using the Sigmoid function, and
graphical analyses of stress-strain and moment-rotation curves. The study developed a
spreadsheet-based tool that enables efficient shear stress-strain prediction by inputting
key structural parameters. This tool offers significant benefits for structural analysis,
design, and the maintenance of buildings, particularly in developing regions.

Keywords: Artificial Neural Networks (ANN), beam-column joint capacity, shear stress-
strain, MATLAB, structural analysis, Levenberg-Marquardt algorithm, TANSIG

activation function, Cook’s Distance Method, predictive modeling, spreadsheet tool.