Innovative Information Systems services have the potential to promote more sustainablebehavior. For these so-called Green Information Systems (Green IS) to work well, they shouldbe tailored to individual behavior and attitudes. Although various theoretical models alreadyexist, there is currently no technological solution that automatically estimates individual’s currentsustainability levels related to their consumption behaviors in various consumption domains(e.g., mobility and housing). The paper aims at addressing this gap and presents the design ofGREENPREDICT, a framework that enables to predict these levels based on multiple features, such asdemographic, socio-economic, psychological, and factual knowledge about energy information. Todo so, the paper presents and evaluates six different classifiers to predict acts of consumption onthe Swiss Household Energy Demand Survey (SHEDS) dataset containing survey answers of 2000representative individuals living in Switzerland. The results highlight that the ensemble predictionmodels (i.e., random forests and gradient boosting trees) and the multinomial logistic regressionmodel are the most accurate for the mobility and housing prediction tasks.