Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/E5F4A228-8B59-4F65-BECC-1C96CD3C070C | ||
| Year | 2023 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Georgios Noikos, "Machine learning to develop a model that will predict early impending sepsis in neurosurgical patients", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023 https://doi.org/10.26233/heallink.tuc.97800 | ||
| Appears in Collections |
As sepsis, we currently define a “life-threatening organ dysfunction caused by a dysregulated host response to infection”. Prevention of sepsis, demands its early pre- diction, a task that has been quite a challenge for the scientific community. With our study, we attempt contributing to this effort, by taking processed, anonymised data, which will be used to build a machine learning predictive model that would predict an upcoming infection, potentially leading to sepsis. Although this model originally takes into consideration, among others, medical measurements of 5 consecutive days, at the end of our study we examine the model’s predictive capacity with a more limited span of days. We even end up predicting based on a single day’s medical measurements, four days prior to infection, obtaining satisfactory results. This goal’s significance is high, since achieving it, would provide the doctors and the nursing staff with some valuable time, constructing an efficient plan to deal with the infection before it causes sepsis. This interval of time could be proven to be decisive about the life of the patient, since sepsis is one of the most frequent reasons for an Intensive Care Unit (ICU) admission and the primary reason for death in the ICU. Data cleaning and pre-processing helped us to feed the best possible dataset to our model, maximizing its predictive capacity for this binary classification problem. Moreover, via the important features of our model, con- clusions could potentially be drawn concerning the relation between some clinical input features and the occurrence of sepsis, leading to an enhanced, data-driven understand- ing of this heterogeneous dysfunction. Early findings indicate efficient classification performance resulting in promising forecasting ability, using various machine learning models, while leaving considerable scope for extending the time between the prediction of the infection and its occurrence.
Copyright © DIAS 2013
