The research and development on machine learning has exploded in recent years due to the revolution of neural networks, which have become the greatest hope in the field of artificial intelligence after the improvements in hardware performance and the availability of high amounts of data. Nevertheless, it appears that advancements have been slowed down recently, probably because neural networks are reaching their precision limits. Moreover, neural networks have an imperfection that has not been remarked frequently: they are a black box in which it is not possible to understand the precision procedure of the network. In consequence, despite their precision, they do not provide us with strong arguments to justify their predictions. Decision trees, on the other hand, are a really interpretable model, though little precise. The objective of the study of this project is the creation of a mixed model between neural networks and decision trees: the deciduous decision tree. This prediction model combines the structure of a decision tree and the predictive capacities of the neural networks, aiming at reaching a custom intermediate point in the dichotomy between precision and interpretability. The results obtained after the analysis of several experiments conducted to the deciduous decision tree reveal that, even with a little loss of precision, the gain of interpretability from the decisions that are made in the tree compensates this loss. Therefore, this new model can be a starting point to make machine learning models be more interpretable, which can be extremely useful in a huge variety of fields, such as medicine, sociology, meteorology, law, etc.