Prediction of urinary tract infection using machine learning methods: a study for finding the most-informative...

Area of Science:

• Medical Informatics
• Computational Biology
• Clinical Diagnostics

Background:

• Urinary tract infections (UTIs) are common and can lead to serious health issues.
• Current diagnostic methods like urine culture are slow and prone to errors.
• There is a need for rapid and reliable UTI diagnostic tools to prevent antibiotic resistance.

Purpose of the Study:

• To identify key predictive variables for urinary tract infection (UTI) using machine learning.
• To evaluate the efficacy of various machine learning models in UTI prediction.
• To establish a more efficient diagnostic approach for UTIs.

Main Methods:

• Employed diverse machine learning algorithms, including classical and deep learning models.
• Analyzed a dataset incorporating urine test results, blood test parameters, and demographic information.
• Utilized an ensemble model combining XGBoost, decision tree, and light gradient boosting with a voting mechanism.

Main Results:

• Identified 18 informative features from urine (e.g., WBC, nitrite), blood (e.g., MPV, lymphocyte), and demographics (age, gender).
• The ensemble model achieved high accuracy (85.64%) and AUC (88.53%) in UTI prediction.
• Significance of gender and age as crucial factors in UTI prediction was demonstrated.

Conclusions:

• Machine learning models show significant potential for accurate and efficient UTI prediction.
• This approach can complement traditional methods, improving diagnostic speed and accuracy.
• The findings support the integration of machine learning in clinical decision-making for UTIs.