SEPSIS HETEROGENOUS CLINICAL DATASET

Description

This dataset consists of 1,700 clinical records created for research on machine learning based early prediction of sepsis from heterogeneous clinical data. The data set includes nine clinical variables that represent vital signs and laboratory measurements, and a binary target variable that represents the presence or absence of sepsis. The predictor variables are Heart Rate, Temperature, Systolic Blood Pressure, Diastolic Blood Pressure, Respiratory Rate, Oxygen Saturation, White Blood Cell (WBC) Count, Lactate Level (mmol/L), and Platelet Count. The target variable Sepsis_Flag is a binary variable that is 0 if the patient does not have sepsis and 1 if the patient has sepsis. There are 1050 records classified as non-sepsis (61.76%) and 650 records classified as sepsis (38.24%). There are missing observations in some clinical variables, especially in the case of Lactate Level, Platelet Count, and WBC Count. The characteristics of this dataset make it appropriate for studying clinical data preprocessing, missing value imputation, class imbalance, feature selection, and supervised machine learning methods for sepsis prediction. The data set was created for the project “Machine Learning Based Early Prediction of Sepsis Using Heterogeneous Clinical Data” at Osun State University, Osogbo, Nigeria. It is designed to enable academic research, experimentation, benchmarking, and development of machine learning models for early sepsis risk prediction. The data set should be used for research and education. It is not intended to be used as a clinically validated diagnostic tool and predictions made using models trained on this data should not replace clinical assessment or clinical decision making. Users of this dataset are requested to cite the dataset and acknowledge the contributors and institution appropriately.

Steps to reproduce

Steps to reproduce 1. Download the Sepsis_dataset_uth.csv file from the Mendeley Data repository. 2. Load the CSV file using Python, R, MATLAB, or another software environment capable of processing tabular data. 3. Inspect the dataset structure and identify the nine predictor variables and the target variable, Sepsis_Flag. 4. The predictor variables are: * Heart Rate * Temperature * Systolic Blood Pressure * Diastolic Blood Pressure * Respiratory Rate * Oxygen * WBC Count * Lactate Level (mmol/L) * Platelet Count 5. Use Sepsis_Flag as the binary target variable, where 0 represents no sepsis and 1 represents sepsis. 6. Inspect the dataset for missing observations. Missing values should be handled using an appropriate imputation strategy, such as mean or median imputation, depending on the analysis. 7. Apply appropriate preprocessing and feature scaling where required by the selected machine learning algorithm. 8. Examine the class distribution and apply an appropriate class-imbalance strategy where necessary. The project methodology identifies SMOTE and other balancing techniques as possible approaches. 9. Divide the dataset into training and testing subsets. The project uses an 80% training and 20% testing strategy. 10. Train the selected machine learning classifiers using the training data. The project evaluates Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Gradient Boosting. 11. Evaluate the trained models using Accuracy, Precision, Recall, F1-score, and ROC-AUC. 12. Generate additional evaluation outputs such as confusion matrices, ROC curves, and feature-importance or model-interpretation results where applicable. 13. Compare the models based on their predictive performance. In the reported experiment, Logistic Regression achieved the strongest overall performance, with 82.06% accuracy, 78.51% precision, 73.08% recall, 75.70% F1-score, and 88.10% ROC-AUC. 14. Use the trained model to estimate sepsis risk for new clinical observations only for research and experimental purposes. The dataset and resulting models are not intended to replace professional clinical diagnosis.

Institutions

• Osun State University

  Osun State, Osogbo
• Kaunas University of Technology

  Kaunas, Kaunas

Categories

Licence