Using Explainable Machine Learning Methods to Predict the Survivability Rate of Pediatric Respiratory Diseases

The mortality rate due to chronic pediatric respiratory diseases is increasing every year and it is important to assess the severity of these diseases. As symptoms of several pediatric respiratory disorders are frequently identical, identification might be difficult due to the ongoing spread of respiratory diseases. Large datasets of clinical variables are analyzed by machine learning (ML) to find patterns and co-relations that human clinicians might not be able to predict immediately. As a result, pediatric respiratory disease severity can be detected more quickly and accurately. The KBest feature selection method is used initially to get the best fifteen features from the dataset. The random forest classifier performed well with the best accuracy of 96% compared to other classifiers. Shapley Additive Values (SHAP), Explain Like I'm 5 (ELI5), QLattice, and Local Interpretable Model-agnostic Explanations (LIME) are four Explainable Artificial Intelligence (XAI) techniques used to interpret model predictions. The most significant attributes were patient transfer to the intensive care unit, Kaliemia, Creatinine Blood Test, Cyanosis, and Natremia. The promising results suggest integrating ML into pediatric respiratory disease diagnosis for predictive accuracy and improved patient outcomes.