Ανάλυση και Επεξεργασία Δεδομένων και Σχεδίαση της Υποκείμενης Γνώσης
Date Issued
June 7, 2021
Type
Πτυχιακή Εργασία
Abstract
In the field of medicine and more specifically in the context of an organized hospital environ-ment, finding knowledge and extracting it is a very critical process. Medicine alone is not able to highlight knowledge, which is essentially a product of experience, but also of the frequency of its occurrence in patient cases. This knowledge, if extracted from the large databases in which it is hidden, is able to strengthen the theoretical medical conclusions and the conclu-sions described in medical textbooks, but also to increase this theoretical knowledge. Experts in the field are able to extract objective results based on their science, using subjective and sub-stantive parameters, such as medical history, test results and more. In this context, ontologies also play an important role. An ontology is the representation of an entity or the concepts of a cognitive area so that it can be processed by the computer. An ontology defines the terms used to describe and represent. Ontologies therefore interpret information and communicate with other tools of logic. With the help of tools, they draw logical conclusions about the concepts that develop within them and the relationships between them.
The purpose of this research is the experimental application of knowledge seeking methods to real medical data. The aim is to seek out and reveal the hidden knowledge, which will be able to be evaluated by health experts. Medical data related to cardiovascular disease were used. As part of the work, three experiments were performed, which showed that the accuracy percent-ages of each algorithm for the individual application of the Attribute Selection filter performed in Experiment 1 are the same as the accuracy rates extracted in Experiment 2 and 3. Experi-ment 3 has almost the same results (accuracy, precision, recall, f-measure) as in Experiment 2.
Reducing features often results in better results because features that are unnecessary or unre-lated to the output value can be ignored. Also, by distinguishing which features are most im-portant for the outcome of a process, intuition can be gained about the real problem, allowing industry experts to deal with it more effectively. When applying the Class Balancer filter, it has been observed that by balancing classes for most algorithms, useful information about occur-rence frequencies is lost.
Subjects