Stabilizing sand with cement is considered to be one of the most cost-effective and useful methods of in-situ soil improvement, and the effectiveness is often assessed using unconfined compressive tests. In certain cases, zeolite and cement blends have been used; however, even though this is a fundamental issue that affects the settlement response of a soil, very few attempts have been made to assess the stress-strain behaviour of the improved soil. Also, the majority of previous studies that predicted the unconfined compressive strength (UCS) of zeolite cemented sand did not examine the effect of the soil improvement variables and strain concurrently. Therefore, in this paper, an initiative is taken to predict the relationships for the stress-strain behaviour of cemented and zeolite-cemented sand. The analysis is based on using the unconfined compression test results and Group Method of Data Handling (GMDH) type Neural Network (NN). To achieve this end, 216 stress-strain diagrams resulting from unconfined compression tests for different cement and zeolite contents, relative densities, and curing times are collected and modelled via GMDH type NN. In order to increase the accuracy of the predictions, the parameters associated with successive stress and strain increments are considered. The results show that the suggested two and three hidden layer models appropriately characterise the stressstrain variations to produce accurate results. Moreover, the UCS values derived from this method are much more accurate than those provided in previous approaches. Moreover, the UCS values derived from this method are much more accurate than those provided in previous approaches which simply proposed the UCS values based on the content of the chemical binders, compaction, and/or curing time, not considering the relationship between stress and strain.Finally, GMDH models can be considered to be a powerful method to determine the mechanical properties of a soil including the stress-strain relationships. The other novelty of the work is that the accuracy of the prediction of the strain-stress behaviour of zeolite-cement-sand samples using the GMDH models is much higher than that of the other models.