The work approaches theoretical and implementation issues of a framework for creating and executing Learning Objects (LOs) where problem-solving tasks are ordered according to the matching of two parameters, both calculated automatically: (1) student skill level and (2) problem solution difficulty. They are formally defined as algebraic expressions. The definition of skill level is achieved through a rating-based measure that resembles the ones of game mastery scales, while the solution difficulty is based on mistakes and successes of learners to deal with the problem. An empirical study based on existing students data demonstrated the suitability of the formulas. Besides, the motivational aspects of learning are considered in depth. In this sense, it is important to propose activities according to the student’s level of expertise, which is achieved through presenting students with exercises that are compatible with the difficulty degree of their cognitive skills. Also, the results of an experiment conducted with four highschool classes using the framework for the domain of logarithmic properties are presented.

O fenômeno da retenção no ensino superior é motivo de preocupações e impactos negativos tanto para a universidade e sociedade. Este trabalho propõe a utilização de técnicas de Mineração de Dados Educacionais e Aprendizagem de Máquina para identificar o perfil de retenção de alunos de graduação em Ciência da Computação da Unioeste, campus de Foz do Iguaçu, aplicando os algoritmos KNN e SVM. Os resultados demonstraram que existe relação entre os dados como sexo, idade e a retenção, sendo útil para a instituição adotar estratégias que visem evitar o baixo rendimento estudantil.