This thesis presents a model for learning mathematics in primary schools, based on mobile learning and data stream mining. The learning contents are presented using educational computer games which allow the students to collaborate through competitive elements and customized social networks. Although personal data were not available, the procedures for anonymizing student-related features were described to prevent identification of an individual through a combination of features. Also, the impact of anonymization on model precision in student classification is analysed. To make it more user-friendly for students, a dedicated model was developed to simultaneously support e-learning in the equivalent way as m-learning, which proved to be the primary choice for students. For ease of understanding and to meet the communication needs between researcher and the experts from the application domain, the requirements and hypotheses of the experts are defined in the codebook. Through interaction with educational computer games in a potentially large number of simultaneous users, the data is streaming into the system, thus making most classical data mining methods unsuitable, especially when there are limitations in scalability. Using of feature selection methods reduces both the dimensions of the datasets and the time required for data processing and diminishes the complexity of the interpretation models without compromising model precision. By applying the methods for data knowledge discovery, based on the data stream mining, student classification models were created. These models enable early detection of gifted students in mathematics, watching streams of data either as stationary or with concept drift. The application of the model proved that there was a statistically significant difference in the style of interaction and the results achieved between gifted mathematics students and other students. A model has also been developed that enables a preliminary grading system based on the assessment of motivation and acquired knowledge. The use of motivational elements in educational computer games and their effect on student engagement were analysed by comparing the results of the control and experimental group. The contents that the students posted on the social network were analysed, whereby the use of deep learning enabled the interpretation of microblogging content.