Identification of novel disease genes is a central part of human and molecular genetics. It is a challenging task that requires large scale case-control studies, which often involve individuals from various human populations. Population structure retrieving and inferences are critical in association studies, in which population stratification can lead to inferential errors. Genotype-based clustering of individuals is an important way of summarizing the genetic similarities and differences between individuals of diverse ancestry. Typical approaches that deal with the problem consist of several time consuming pre-processing steps, such as variant filtering, LD-pruning and dimensionality reduction of genotype matrix. However, these algorithms need to be well-tuned to complete the learning process in the best possible way and considerably low computation time. Traditional parallelization strategies cannot scale with variable data sizes at runtime. To address this issue, we have developed a framework using R programming language for fine-tuning the quality-control and the classification model parameters, and a distributed computing framework for scalable stratification analysis in Apache Spark that is more effective in processing of datasets containing large number of observations (i.e. number of samples > 10, 000). Performed tests confirmed the efficiency and scalability of the presented approach. The implemented tools can also be applied to any variant data set, including data from large scale sequencing projects or custom data sets from clinical laboratories.