How can we validate a given Multiple-Criteria Decision-Making (MCDM) approach and compare it to other MCDM methods? This question is not addressed in MCDM literature. This paper presents a novel out-of-sample approach using in-sample (e.g., 70%) data to assess the parameters of a given MCDM model and validate the model using out-of-sample (e.g., 30%) data. MCDM models are ranked based on their accuracies of predicting the out-of-sample data using several randomly selected replicas. We develop a new class of MCDM models based on Geometric Dispersion Theory (GDT) that was recently developed for decision-making under risk. MCDM-GDT is based on a convex combination of maximizing an additive utility function (an arithmetic function) and a multiplicative utility function (a geometric function); this MCDM-GDT function is non-additive and non-linear in criteria and in weights of importance of criteria. A special case of MCDM-GDT is a highly effective concave utility function using k+1 parameters where k is the no. of criteria. We apply MCDM approach for selection of the best wind farm location in Saudi Arabia using 30 expert decision makers (DMs). Then we provide a detailed comparison of several well-known MCDM methods and MCDM-GDT using the out-of-sample approach. The results indicate that most of MCDM methods, except MCDM-GDT, have poor predictive performances. We also develop a new model for group decision-making using a new nonlinear aggregation of the ratings of DMs based on MCDM-GDT; we show that this model has advantages to using the commonly used weighted average rating of DMs.