In academia, geopolymers/alkali activated materials are becoming increasingly popular, as researchers are exploring a substitute for Portland cement, that is cost-effective, merchantable, and potent. One such potential-seeking sector is the use of geopolymers in the oil and gas industry for well cementing zonal isolation operations. Though it is yet, to be implemented in the field, there has been a few researchers, although not so many in numbers, that conducted geopolymer lab-tests (controlled and expensive environment). Pushed by the fact that any product must be tested under a vast space of external conditions, before being commercialized; the authors wish to address the gap, by applying a variety of prediction models, with the aim to produce accurate results, while relying on a limited set of experimental data. Binary/multi thresholds classification (logistic/probit, decision tree, random forest, SVM), as well as regression and continuous models (multi linear regression, neural networks, among others) are used to predict an important property of the geopolymers (pumpability). This is important, as despite the proven strong performance of such models in other areas, the novelty of the product/subject, uniqueness/insufficiency of the data and the high sensitivity in the behavior of the geopolymers (especially for the pumpability property) subject to slight changes to the chemical mixture design, accurate results and validation is yet to be tested, and most importantly, the ability of the models to generalize. The study uncovers that Decision Tree model provides a simple and intuitive way to understand the behavior of the geopolymers, subject to a variety of external conditions, and can in fact, be used by future users, to accurately predict the pumpability conditions. Although present, inaccurate/poor predictions (false positive) with high operational risk (defined as a geopolymer not reaching its polycondensation phase) have very low probability of occurrence.