Objective:

To evaluate the performance of machine learning models, including SHAP analysis, in predicting the risk of atelectasis in children with Mycoplasma pneumoniae pneumonia.

Key Findings:

• The neural network model achieved the best performance with an AUC of 0.89 and accuracy of 0.82, as identified through SHAP analysis.
• KNN model showed comparable performance (AUC = 0.88), while SVM had the highest specificity (0.87).
• Key variables influencing predictions included neutrophil percentage, serum amyloid A, and C-reactive protein, as highlighted by SHAP analysis.

Interpretation:

The study demonstrates the effectiveness of machine learning models, particularly neural networks, in predicting atelectasis risk in pediatric MPP, providing clinicians with interpretable tools for early identification and management, which can significantly improve patient outcomes.

Limitations:

• The study is retrospective and may be subject to biases inherent in such designs, potentially affecting the reliability of the findings.
• The sample size, while substantial, may limit generalizability to broader pediatric populations, necessitating further validation in diverse cohorts.

Conclusion:

Machine learning models, especially neural networks, can effectively predict atelectasis risk in children with MPP, aiding in early intervention and management.