<p>Built-up battened columns are frequently utilized in civil engineering structures; however, their seismic response has not been extensively studied, particularly when subjected to excitation along their weak axis. This research simulated 192 built-up battened columns with varying batten spacing, chord distances, axial forces, and batten thicknesses using ABAQUS software, applying similar quasi-static cyclic excitation along both weak and strong axes. Furthermore, the effectiveness of neural networks in predicting the bending capacity of these columns was explored. Results revealed that columns excited along the strong axis exhibited a larger plastic zone at the first panel, a higher effective yield strength and stiffness, and a more rapid strength degradation rate compared to those excited along the weak axis. Additionally, the displacement corresponding to their ultimate load was smaller. Moreover, the bending capacities of all columns were less than their plastic moments, regardless of batten spacing, chord distance, and excitation direction. All columns demonstrated a displacement ductility ratio of less than two, indicating a force-controlled cyclic response. The findings of this study suggest that the use of built-up battened columns should be avoided in regions with high seismic activity, as they display limited ductility and overstrength ratios. Furthermore, this study indicates that the equations used to calculate the ultimate bending capacity of built-up columns require modifications, as they overestimate the values. Additionally, it is shown that feed-forward neural networks can effectively estimate the bending capacity of built-up battened columns while considering the effects of axial force and batten spacing.</p>