Methods for Removing Non-Interesting Itemsets when Mining Electronic Healthcare Records

The usage of data mining techniques in healthcare has exponentially increased in the last years. Analyzing the huge amount of data that is nowadays produced by healthcare systems can lead to the extraction of useful and interesting informations about patients and diseases, which can be exploited to improve medical research and knowledge. Understanding how diseases and other characteristics of a patient are interrelated is a crucial point because it can help healthcare specialists to focus only on important factors when addressing cures for a given clinical case. Frequent itemset mining techniques are widely used for this purpose, but they can lead to the retrieval of too many redundant or not interesting pieces of information. In this project we study and report performances of a measure proposed to remove redundant and irrelevant rules from data and suggest an approach to unveil the main comorbidities for a given disease, along with the possibility to use the latter results to further filter not interesting informations. Results show the effectiveness of the two studied methods as also proved by the main literature that was reviewed during the project, even if we suggest the collaboration with healthcare specialists in order to get more relevant outcomes.