Recent years have seen the rise and evolution of automated systems capable of learning and performing tasks with minimal or no human intervention. This field, which took the name Artificial Intelligence (AI), comprises several strategies to enable machines to perform different tasks. Research in this field experienced a rush for performance, striving to obtain the best-performing model possible, which led to Deep Learning (DL). Unfortunately, focusing only on performance compromises other aspects, such as the transparency of the model. Most ML and DL models also struggle with small datasets, missing information, and the presence of noise in the data. These limitations can impair their performance despite their strong theoretical foundations. We focus on these constraints, providing a framework that can perform under such conditions while being inherently explainable. The main application of this research is healthcare, due to the typical limitations in the quality and quantity of data inherent to this field, especially for rare conditions. We explored a novel approach to classification named CACTUS<\/strong> (the Comprehensive Abstraction and Classification Tool for Uncovering Structures<\/a><\/strong>), which shows promising results in overcoming some limitations in ML, identifying and addressing open challenges in classification, with an emphasis on AMD diagnosis.<\/p>\n\n\n\nAbout the author:<\/h2>\n\n\n\n