Authors
- Jelena Musulin, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Ivan Lorencin, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Nikola Anđelić, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Sandi Baressi Šegota, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Daniel Štifanić, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Vedran Mrzljak, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
- Elitza Markova-Car, University of Rijeka, Department of Biotechnology, Rijeka, Croatia
- Zlatan Car, University of Rijeka, Faculty of Engineering, Rijeka, Croatia
Article type:
Original Scientific Paper
Abstract:
Sepsis is the most severe consequence of bacterial infections, in other words, a potentially life-threatening condition. During infection, some components of the innate immune response can, under certain circumstances, cause multiple organ dysfunction and tissue damage. Fortunately, timely treatment can prevent the consequences.
The algorithms used to solve the problem of predicting sepsis occurrence are Support-Vector Machine (SVM), K-Nearest Neighbors (KNN), and Artificial Neural Network (ANN). The dataset used for this research consists of 23 different medical parameters for each patient, which is a total of 2277 data points.
For each of the aforementioned algorithms, different combinations of parameters were examined as well as different types of kernel functions (SVM), architectures (ANN), and number of nearest neighbors (KNN). The experimental results showed that the ANN approach achieves the highest AUC value (0.992) compared to the other approaches like SVM and KNN.
Keywords:
Sepsis, Artificial Neural Networks, Support Vector Machines, K-Nearest Neigh- bors