Rough sets based on reducts of conditional attributes in medical classification of the diagnosis status

Rough sets constitute helpful mathematical tools of the classification of objects belonging to a certain universe when dividing the universe in two collections filled with sure and possible members. In this work we adopt the rough technique to verify diagnostic decisions concerning a sample of patients whose symptoms are typical of a considered diagnosis. The objective is to extract the patients who surely suffer from the diagnosis, to indicate the patients who are free from it, and even to make decisions in undefined diagnostic cases. We also consider a decisive power of reducts being minimal collections of symptoms, which preserve the previous classification results. We use them in order to minimize a number of numerical calculations in the classification process. Finally, we test influence of symptom intensity levels on the diagnosis indisputable appearance to select these levels that are expected to be found in patients suffering from the considered diagnosis. The presence or the absence of these symptom levels in the patients allow us to add complementary remarks to earlier classification effects making them even more readable.