Conventional reinforcements used for soil strengthening are planar in nature. The concept of three dimensional reinforcements is gradually gaining popularity. Such reinforcements possess the inherent advantages of the conventional planar reinforcements. The additional protrusions on their surface adds to the passive resistance to shear deformation. In this study, a series of drained and undrained triaxial compression tests were performed on unreinforced and reinforced sands. The variables considered include the gradation of soil in terms of the effective grain size, volume ratio of reinforcement, reinforcement orientation, spacing and confining pressure. The results obtained were used to train two models, according to the testing conditions. The first model was used for predicting the peak shear stress in dry samples and the second predicted the peak shear in saturated samples. In each of these models, the output layer consisted of a single node i.e., peak shear. The results show that ensuring a proper training and a learning algorithm can help in developing an Artificial Neural Network (ANN) model that could serve as an effective tool in predicting shear strength improvement of sands reinforced with multi-directional inclusions. This would in turn help practising engineers to estimate the improvement in shear strength parameters before these 3D reinforcements are actually added to the soil.