A soil investigation work was conducted at Shalimar, Howrah, with fourteen (14) numbers of boreholes. The subsoil parameters were determined by in situ laboratory tests and recorded as borehole profile. However, some of the crucial parameters like shear strength of soil in a borehole profile appear to be missing at some depths presumably for avoiding extensive laboratory tests. On the basis of the available laboratory test data, the present study attempts to develop machine learning-based regression models for estimating any depth-specific cohesion value as one of the presumably missing crucial shear strength parameters of borehole profile depending on other available subsoil parameters of the profile at same depth. The cohesion values estimated by the models are validated against the actually available cohesion values logged in the borehole profiles. The comparative analysis based on the experimental results indicates that the ANN-based Convolutional Regression model (3 hidden layers and 64 neural units/layer) outperforms the popular linear regression model as it achieves a promising estimation accuracy with 93% R-squared score.