A Standardized Data Mining Method in Healthcare: a pediatric intensive care unit case study
MetadataShow full item record
The growth of available data in the healthcare led to numerous data mining projects being launched over the years, that revolves around knowledge discovery. In spite of this, the medicine domain experiences several challenges in their quest of extracting useful and implicit knowledge due to its inherent complexity and unique characteristics, as well as the lack of standards for data mining projects. Hence, the aim of this research is to bring some standardization in data mining processes in the healthcare based on the Cross-Industry Standard Process for Data Mining (CRISP-DM) method. The CRISP-DM is widely adopted in various industries and is suitable as a base method on which enhancements can be made in order to bring domain specific standardizations. This proposed method which is named MSP-DM was evaluated by domain experts from the UMC and UU. Additionally, these expert interviews were conducted in identifying any missed method fragments that were not captured during the case study or mentioned in the literature, as well as evaluting the found method fragments. During the course of the case study, one of the provided projects was successfully completed and implemented, as for the second project insight was gained about the possibilities of predictive modeling. Moreover, during the expert sessions and the case study, a high emphasis was given to more involvement of clinical professionals and domain experts during a data mining project, i.e. in the selection of parameters, modeling, and evaluation. The clinical staff is usually unacquainted with the concept of data mining, which can create a gap between the researcher performing the analysis and the (medical) domain experts. Similarly, not involving clinical practitioners in the data mining project could lead to a failure to adopt a certain technology or analysis result, because the clinical practitioners could feel surpassed not being consulted or involved in the process. In addition, for researchers that are unfamiliar with the medicine domain it is essential to interact with clinical professionals in order to attain a sufficient understanding of the domain, which will eventually help in comprehending the problem, data, and objectives. Hence, a collaboration is required in mitigating this problem through their (clinical practitioners) provided input that can determine relevant outcomes and issues, which will lead to better analysis and easy implementation of the outcomes that are found. Likewise, practical activities and concepts were found that were missing in the original method. For this reason, these and other findings were incorporated in the MSP-DM, which proved to be viable during the case study. In consideration with the results, the created method provides an extension of the CRISP-DM tailored for the healthcare that includes the current challenges of data mining projects which may be extended to comprise processes relevant to other domains as well.