Analysis of Imputation Techniques and Classification Types in the Improvement of Early Diagnosis of Infectious Diseases in Elderly People

Early diagnosis of infections in institutionalized individuals is vital for improving quality of life, reducing mortality, and preventing healthcare system overload. Vital signs such as electrodermal activity, blood pressure, and body temperature can predict infections using artificial intelligence. However, data quality issues, such as missing records and outliers, complicate the model accuracy. This study explores advanced imputation methods (mean, k-Nearest Neighbors, and Multiple Imputation by Chained Equations) to improve infectious disease detection. Using XGBoost, binary classification (infection/no infection) significantly outperforms previous results, achieving an F1-score of 0.784, up from 0.664. For multiclass classification, the F1-score reaches 0.874, surpassing the previous 0.776.