Concrete-filled steel tubes (CFSTs) are widely used in engineering structures due to their excellent mechanical properties and economic benefits. This study focused on the construction of artificial neural network (ANN) models with high prediction capabilities and prediction accuracies that could predict the axial compression load capacities of short CFST columns using machine learning methods. A database was created by searching literature published over the past 40 years regarding circular-CFST bearing-capacity testing. Three ANN models with different input parameters were developed, and used the Whale Optimization Algorithm to optimize the network weights and thresholds, the core idea of which comes from the humpback whale's special bubble net attack method. Then, the predictions of the proposed machine learning models were also compared with the theoretical values produced by the formulas proposed in existing codes. The results show that the ANN models had higher accuracies and a wider application range than the existing code models. Based on the Garson's algorithm, we perform parameter sensitivity analysis on the network model to enhance the interpretability of the neural network model. Finally, a graphical user tool is built to make the strength of CFST can be predicted quickly.

钢管混凝土（CFST）因其优异的力学性能和经济效益而广泛应用于工程结构中。本研究的重点是构建具有高预测能力和预测精度的人工神经网络（ANN）模型，可以使用机器学习方法预测短钢管混凝土柱的轴向压缩载荷能力。通过检索过去 40 年来发表的有关圆形钢管混凝土承载能力测试的文献，创建了一个数据库。开发了三种不同输入参数的ANN模型，并使用鲸鱼优化算法来优化网络权重和阈值，其核心思想来自座头鲸特殊的气泡网攻击方法。然后，还将所提出的机器学习模型的预测与现有代码中提出的公式产生的理论值进行比较。结果表明，人工神经网络模型比现有的代码模型具有更高的精度和更广泛的应用范围。基于Garson算法，我们对网络模型进行参数敏感性分析，以增强神经网络模型的可解释性。最后，构建了图形用户工具，使得CFST的强度可以快速预测。