This paper proposes a machine learning framework to discover and analyse the drawing strategies observed in the embedded figure tests. This type of test is used in psychology, psychiatry, and neurology to detect deviations in cognitive development. Unlike many other drawing tests, embedded tests did not receive much attention from the digitisation perspective. At the same time, using the tablet PC to perform these types of tests may allow one to extract information that is unavailable when a test is performed with pen and paper. The working hypothesis of the present investigation is that the strategies used to trace elements of the reference figure embedded in another figure should be different between groups of subjects with different levels of literacy. To confirm this hypothesis, a tablet PC equipped with custom software was used to perform the tests. Then, different strokes of the test were extracted and classified using deep learning machinery. The association pattern mining approach was then used to extract and analyse peculiar patterns for groups with different literacy levels. The frequencies of the most common stroke subsequences were used as features of the decision tree classifier to estimate the literacy class of previously unseen tests. The classifier has achieved a precision of 0.88, recall of 0.83, and accuracy of 0.83.