In the complex study to obtain indicators in the autism spectrum disorder it is very common to perform many and very complex tasks. Often, these tasks require the completion of a series of forms and surveys that are even more complex and tedious, which means that the accuracy of the reports is not always satisfactory. In this paper, we propose a general architecture based on machine learning techniques and data mining for prediction of the main indicators in the diagnosis and intervention of the autistic spectrum disorder. The main idea of this approach is to replace those print documents by mobile tests, tablet or smartphones tests through games, store them in databases and analyse them. Furthermore, very often these last two steps are not undertaken with the lack of quantitative and qualitative analysis that could be generated. Finally, the presented architecture is oriented to data collection with the objective of the creation of large specialized datasets. © 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.