Abstract: Accurate anticipation of soil compression modulus is crucial for geological hazard assessment, particularly in cases where cone penetration test (CPT) data are scarce. This study proposes a novel approach, combining a convolutional neural network (CNN) model and Kriging interpolation, to effectively predict soil compression modulus and visually display its spatial distribution characteristics. The convolutional neural network model, with its architecture and parameters determined through training, learned from sample data to investigate the implicit correlation between CPT parameters and the compression modulus. It utilized the random field generated by Kriging interpolation as input to predict the compression modulus of the site. The results showed that Kriging interpolation was able to accurately forecast cone tip resistance and sleeve friction. The variability of these parameters was pronounced with depth but less significant horizontally. In the CNN model employing small convolutional kernels, short strides, and a multi-feature extraction architecture, the predicted random field of compression modulus aligned well with empirical data. The predicted compressive modulus, validated at a specific location, exhibited a correlation coefficient (R²) of 0.787. Therefore, the predicted compression modulus is quite representative of the site, and this consistency can serve as a reference for settlement computations and risk evaluations.