Peat lands cover many areas around the world in several countries such as Canada, the United States, Malaysia, Finland and others. Generally, construction over peat is a risky business due to its inherent engineering and physical characteristics that make such construction difficult and, in some cases, impossible to perform. Typical characteristics of peats include high organic content, low shear strength, low bearing capacity, high compressibility and high moisture content. Engineers and researchers involved in construction over peat have opted, throughout the years, to strengthen peat subgrades by using different types of stabilization techniques to enhance their field performance. Several researchers have undertaken research studies to investigate the reinforcement effects of various geosynthetics. Peats, however, were less investigated especially with regard to geogrid reinforcement. High spatial variability characterizes soil properties in general and peats, which makes predicting their in-situ properties difficult to achieve. This study aims to evaluate the applicability of using the Group Method Data Handling Neural Networks (GMDH-NN) in predicting the base layer thickness in the design of unpaved roads on Malaysian peats. Giroud and Han’s model is used as a benchmark in this evaluation. Conclusions and recommendations are drawn to help engineers and practitioners in the decision process in current and future unpaved road base applications.