Concrete is the most commonly used building construction material in modern infrastructure, and shear strength is a critical characteristic that defines its performance. This paper discusses using a Convolutional Neural Network (CNN) to improve the accuracy of concrete shear strength prediction by using artificial intelligence technology. In this study, 994 data points are divided into three groups: 795 training data points, 149 validation data points, and 50 testing data points. For CNN model generalisation, each data point was normalised (0–1). The CNN architecture was designed using five convolutional layers with 2 × 2 kernels, batch normalisation, ReLU, max pooling, and dropout, followed by a fully connected and regression layer. The data was trained using RMSProp, with a mini-batch size of 8, 120 epochs, and periodic validation on the validation dataset. The results from the CNN model indicate that during prediction, linear regression after data normalization achieved R2 = 0.9911, a mean error of 0.41236, and an RMSE of 0.018734. The predicted concrete shear strength is 0.231, i.e., above the SNI and ACI Code estimates of 0.17. The parameters strongly influencing concrete shear strength are rho and dimensions a, d, and h. Moderately influential parameters include Vtest, \({\text{f}}_{{\text{c}}}^{\prime}\) , As, and fy.