Supervised machine learning for early predicting the sepsis patient: modified mean imputation and modified chi-square feature selection

• Published in: Multimedia tools and applications Vol. 80; no. 13; pp. 20477 - 20500
• Main Authors: Shrestha, Ujjwol, Alsadoon, Abeer, Prasad, P. W. C., Al Aloussi, Sarmad, Alsadoon, Omar Hisham
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2021; Springer Nature B.V
• Subjects: Accuracy; Chi-square test; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Emergency medical services; Feature selection; Machine learning; Missing data; Multimedia Information Systems; Sepsis; Special Purpose and Application-Based Systems; Specificity; Sensitivity; Supervised learning; Prediction; Machine learning; Sepsis; Electronic health records
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-021-10725-2

Sepsis is a typical and significant emergency in medical clinics comprehensively. A creative and possible instrument for identifying sepsis stays elusive. Supervised models can identify potential clinical factors and give a more accurate prediction than the existing benchmark rule-based tools. This research aims to increase the sensitivity to accurately predict the sepsis patient. The proposed system consists of the mean imputation and chi-square technique to replace the missing features and feature selection, respectively. All datasets are fed into the chi-square technique for feature selection by measuring how expectations compare to actual observed data. The essential missing data are then replaced using the mean-imputation method by calculating the mean value of the available data. Finally, the selected features are used as an input to the supervised machine learning model for the classification of sepsis patient. The results of accuracy and processing time are obtained by using different datasets. The results show that the proposed solution achieves better classification performance in different data scenarios and different review types. The proposed solution provides a classification accuracy of 97.67% against the current accuracy of 91.12% on average. It also provides a processing time of 29.1 milliseconds against the current processing time of 32.8 milliseconds on average. The proposed system is focused on the feature selection process that is involved in the machine learning model. Finally, this study solves the issue of model overfitting with supervised machine learning.